test(milestones): add full milestone run tests

Add comprehensive tests that run each milestone script fully:
- Tests all 6 milestones (01-06) with actual training
- Verifies correct outputs and accuracy thresholds
- Marked as @pytest.mark.slow for release validation
- Suitable for e2e testing, not regular CI

These tests validate the complete educational experience works end-to-end.

tinytorch/tests/milestones/test_milestones_run.py

@@ -0,0 +1,196 @@
+"""
+Milestone Full Run Tests
+========================
+
+These tests run each milestone script fully to verify the complete
+educational experience works end-to-end.
+
+This is Option C: Full run (~10-15 minutes total)
+- Runs each milestone with actual training
+- Verifies outputs are correct (accuracy thresholds, etc.)
+- Suitable for release validation, not regular CI
+
+Usage:
+    pytest tests/milestones/test_milestones_run.py -v
+    pytest tests/milestones/test_milestones_run.py -v -k "milestone_01"
+"""
+
+import subprocess
+import sys
+import os
+import pytest
+from pathlib import Path
+
+
+# Get the tinytorch root directory
+TINYTORCH_ROOT = Path(__file__).parent.parent.parent
+
+
+def run_milestone(milestone_id: str, part: int = None, timeout: int = 300) -> tuple[int, str, str]:
+    """
+    Run a milestone via tito CLI and capture output.
+
+    Args:
+        milestone_id: Milestone ID (01-06)
+        part: Optional part number for multi-part milestones
+        timeout: Timeout in seconds (default 5 minutes)
+
+    Returns:
+        (return_code, stdout, stderr)
+    """
+    # Use the bin/tito script directly
+    tito_script = TINYTORCH_ROOT / "bin" / "tito"
+
+    cmd = [
+        str(tito_script),
+        "milestone", "run", milestone_id,
+        "--skip-checks"  # Skip prerequisite checks since we're testing
+    ]
+
+    if part is not None:
+        cmd.extend(["--part", str(part)])
+
+    env = os.environ.copy()
+    env["TITO_ALLOW_SYSTEM"] = "1"  # Allow running without venv
+    env["PYTHONPATH"] = str(TINYTORCH_ROOT)
+
+    # Auto-answer prompts by providing 'n' to stdin (decline syncing achievements, etc.)
+    result = subprocess.run(
+        cmd,
+        cwd=TINYTORCH_ROOT,
+        capture_output=True,
+        text=True,
+        timeout=timeout,
+        env=env,
+        input="n\nn\nn\n"  # Answer 'n' to any prompts
+    )
+
+    return result.returncode, result.stdout, result.stderr
+
+
+class TestMilestoneRuns:
+    """Test that all milestones run successfully and produce correct output."""
+
+    @pytest.mark.slow
+    def test_milestone_01_perceptron(self):
+        """Milestone 01: Perceptron (1958) - Forward pass with random weights."""
+        returncode, stdout, stderr = run_milestone("01", timeout=60)
+
+        # Should complete (even with errors in output, the script should finish)
+        assert returncode == 0, f"Milestone 01 failed:\nstdout: {stdout}\nstderr: {stderr}"
+
+        # Should show the perceptron architecture
+        assert "Perceptron" in stdout or "perceptron" in stdout.lower()
+
+        # Should mention random weights (this is forward-pass only)
+        assert "random" in stdout.lower() or "Random" in stdout
+
+    @pytest.mark.slow
+    def test_milestone_02_xor_crisis(self):
+        """Milestone 02: XOR Crisis (1969) - Demonstrates XOR problem."""
+        returncode, stdout, stderr = run_milestone("02", timeout=60)
+
+        assert returncode == 0, f"Milestone 02 failed:\nstdout: {stdout}\nstderr: {stderr}"
+
+        # Should mention XOR
+        assert "XOR" in stdout or "xor" in stdout.lower()
+
+        # Should show the limitation (can't solve XOR with single layer)
+        assert "75%" in stdout or "50%" in stdout or "cannot" in stdout.lower() or "limit" in stdout.lower()
+
+    @pytest.mark.slow
+    def test_milestone_03_mlp_revival(self):
+        """Milestone 03: MLP Revival (1986) - Solves XOR and trains on digits."""
+        returncode, stdout, stderr = run_milestone("03", timeout=180)
+
+        assert returncode == 0, f"Milestone 03 failed:\nstdout: {stdout}\nstderr: {stderr}"
+
+        # Should solve XOR with 100% accuracy
+        assert "100" in stdout and ("XOR" in stdout or "accuracy" in stdout.lower())
+
+        # Should train on digits
+        assert "digit" in stdout.lower() or "Digit" in stdout
+
+    @pytest.mark.slow
+    def test_milestone_04_cnn_tinydigits(self):
+        """Milestone 04: CNN Revolution (1998) - TinyDigits (default, no download)."""
+        returncode, stdout, stderr = run_milestone("04", timeout=180)
+
+        assert returncode == 0, f"Milestone 04 failed:\nstdout: {stdout}\nstderr: {stderr}"
+
+        # Should use TinyDigits (not CIFAR)
+        assert "TinyDigits" in stdout or "tinydigits" in stdout.lower() or "8x8" in stdout
+
+        # Should achieve reasonable accuracy (>70%)
+        # Look for accuracy numbers in the output
+        import re
+        accuracy_matches = re.findall(r'(\d+(?:\.\d+)?)\s*%', stdout)
+        if accuracy_matches:
+            # Get the highest accuracy mentioned (likely final test accuracy)
+            accuracies = [float(a) for a in accuracy_matches if float(a) <= 100]
+            if accuracies:
+                max_accuracy = max(accuracies)
+                assert max_accuracy >= 70, f"CNN accuracy too low: {max_accuracy}%"
+
+    @pytest.mark.slow
+    def test_milestone_05_transformer(self):
+        """Milestone 05: Transformer Era (2017) - Sequence reversal with attention."""
+        returncode, stdout, stderr = run_milestone("05", timeout=180)
+
+        assert returncode == 0, f"Milestone 05 failed:\nstdout: {stdout}\nstderr: {stderr}"
+
+        # Should mention attention/transformer
+        assert "attention" in stdout.lower() or "transformer" in stdout.lower()
+
+        # Should achieve good accuracy on reversal task (>90%)
+        import re
+        accuracy_matches = re.findall(r'(\d+(?:\.\d+)?)\s*%', stdout)
+        if accuracy_matches:
+            accuracies = [float(a) for a in accuracy_matches if float(a) <= 100]
+            if accuracies:
+                max_accuracy = max(accuracies)
+                assert max_accuracy >= 90, f"Transformer accuracy too low: {max_accuracy}%"
+
+    @pytest.mark.slow
+    def test_milestone_06_mlperf(self):
+        """Milestone 06: MLPerf Benchmarks (2018) - Optimization techniques."""
+        returncode, stdout, stderr = run_milestone("06", timeout=180)
+
+        assert returncode == 0, f"Milestone 06 failed:\nstdout: {stdout}\nstderr: {stderr}"
+
+        # Should mention optimization techniques
+        assert any(term in stdout.lower() for term in [
+            "quantiz", "compress", "cache", "kv", "speedup", "accelerat"
+        ])
+
+        # Should show compression ratio (4x for INT8)
+        assert "4" in stdout and ("compress" in stdout.lower() or "×" in stdout or "x" in stdout.lower())
+
+
+class TestMilestoneSequence:
+    """Test that milestones can be run in sequence (simulates student journey)."""
+
+    @pytest.mark.slow
+    @pytest.mark.parametrize("milestone_id", ["01", "02", "03", "04", "05", "06"])
+    def test_milestone_completes(self, milestone_id):
+        """Each milestone should complete without errors (or with expected bonus challenge failures)."""
+        returncode, stdout, stderr = run_milestone(milestone_id, timeout=300)
+
+        # Milestone 05 has bonus challenges that may fail - the core reversal task passing is sufficient
+        # The individual test_milestone_05_transformer test validates the actual learning objective
+        if milestone_id == "05":
+            # Check that reversal (Challenge 1) passed - this is the core learning objective
+            assert "Reversal" in stdout and "PASSED" in stdout, (
+                f"Milestone 05 reversal challenge should pass:\n"
+                f"stdout: {stdout[-2000:] if len(stdout) > 2000 else stdout}"
+            )
+            return  # Bonus challenges may fail, that's OK
+
+        assert returncode == 0, (
+            f"Milestone {milestone_id} failed with return code {returncode}\n"
+            f"stdout: {stdout[-2000:] if len(stdout) > 2000 else stdout}\n"
+            f"stderr: {stderr[-500:] if len(stderr) > 500 else stderr}"
+        )
+
+        # Should show achievement unlocked
+        assert "MILESTONE" in stdout.upper() or "Achievement" in stdout or "✅" in stdout