feat: support postselection masks in detector sampler

[CodeMaverick2]

Summary

Closes - #41

• Add optional postselection_mask to CompiledDetectorSampler.sample (feat: Support postselection masks #41)
• Compute direct detector bits in NumPy and skip JAX for shots discarded by direct postselected detectors
• Buffer survivors into fixed-size JAX batches with padding on the final batch
• Apply detector reference XOR before postselection checks and on returned detector columns
• Fully-direct circuits continue using the existing NumPy fast path

Test plan

• uv run pytest (1102 passed)
• uv run pre-commit run --all-files
• New unit tests in test/unit/test_postselection.py (18 tests)