Replace obsolete Boost seed-invariance test

[OVVO-Financial]

Motivation

• Remove an ad-hoc pytest collection hook and a separate reproducibility test and restore the intended, direct parity check in the original Boost parity test to keep parity logic centralized and Ruff-clean.
• Replace the outdated seed-invariance regression guard with a focused same-seed reproducibility test to ensure reproducible delegated NNS.stack estimates.

Description

• Edited tests/parity/test_boost.py to replace the obsolete test_nns_boost_ivs_test_none_is_seed_invariant with a parametrized same-seed reproducibility test test_nns_boost_ivs_test_none_is_seed_reproducible that matches the requested implementation and asserts identical results when reusing a random_seed.
• Updated the stale comment in tests/parity/test_boost.py to note that the default seed 123 is used to match R NNS.boost and its delegated NNS.stack folds.
• Removed the separate workaround artifacts by ensuring the standalone reproducibility test and parity conftest.py collection hook are not used (the branch contains only the direct edit to the original Boost parity test).
• Did not modify any production code (src/nns/boost.py, src/nns/stack.py) or configuration files listed as off-limits.

Testing

• Ran python -m ruff check . --fix and python -m ruff check ., and both completed with no remaining Ruff violations.
• Attempted NNS_R_CACHE_ONLY=1 python -m pytest -q tests/parity, python -m pytest -q tests/invariants, and python -m pytest -q tests/property, but these runs failed locally due to missing runtime test dependencies (ModuleNotFoundError: No module named 'numpy').
• Ran python -m mypy and python -m build, which could not complete in this environment due to missing build/runtime dependencies and toolchain components (NumPy/matplotlib stubs and the build package).

---

Codex Task