refactor: consolidate sandbox into competition/ package, remove redundancies

src/sandbox/competition/__init__.py

@@ -0,0 +1,38 @@
+"""
+KernelGenBench Competition Anti-Cheat Package.
+
+7-layer sandbox for detecting cheating in Triton kernel submissions.
+
+Layers:
+  1. File System Isolation   — cache_isolator.py
+  2. Environment Variables   — applied in orchestrator
+  3. Import Hook Sandbox     — import_hook.py
+  4. CUDA Layer Protection   — cuda_protector.py
+  5. Random Shapes           — shape_generator.py
+  6. Process Isolation       — process_isolator.py (via multiprocessing)
+  7. Timing Validation       — timing_validator.py
+
+Usage:
+    from sandbox.competition import run_check, CheckConfig
+
+    result = run_check(
+        kernel_path="path/to/kernel.py",
+        generate_inputs=lambda: (torch.randn(128, 512, device='cuda'),),
+    )
+    print(result.passed)   # True if no cheating detected
+    print(result.reason)   # Explanation
+"""
+
+from .orchestrator import run_check, CheckConfig, CheckResult
+from . import cache_isolator
+from . import import_hook
+from . import cuda_protector
+from . import shape_generator
+from . import timing_validator
+from . import process_isolator
+
+__all__ = [
+    "run_check",
+    "CheckConfig",
+    "CheckResult",
+]