fix(baseline): widen pattern file lists to accept README.rst / .txt / no-ext variants

[mlieberman85]

Summary

Follow-up to #402 (currently in review as #406). Where #406 was TOML-only inside llm_eval blocks, this widens the sibling pattern handler's files list on the same 5 controls that gate an llm_eval pass, so the deterministic tier resolves PASS/FAIL conclusively on more repos rather than falling through to the LLM tier.

Motivation

The bug pattern that #402 fixed downstream: a Sphinx-based Python project that ships README.rst but no README.md would have the pattern.files list scan zero on-disk candidates, return INCONCLUSIVE via _regex_no_files_result, and the pipeline would fall through to llm_eval. #406 makes that llm_eval fallback useful (real file content); this PR makes the fallback fire less often by covering rst/txt/no-ext README on the deterministic tier.

Controls widened

Control	Added
OSPS-DO-03.02	README.rst, README.txt, README
OSPS-DO-04.01	README.rst, README.txt, README, docs/support.md
OSPS-DO-05.01	README.rst, README.txt, README
OSPS-SA-01.01	rst variants for ARCHITECTURE.md, DESIGN.md, README.md, docs paths
OSPS-SA-02.01	rst variants for API.md, README.md, USAGE.md, docs/api.md

_regex_match_files reads each listed file in full; the additions stay well under the handler's _FILE_SCAN_LIMIT = 100 cap and the per-file 2000-byte truncation.

Test plan

• tests/darnit_baseline/controls/test_pattern_file_variants.py (new, 8 parametrized tests): asserts README.rst is present on every gated control's pattern list; asserts README.txt / README (no ext) on the DO-* controls where they were added.
• pytest tests/darnit_baseline/ -q -> 877 pass, 7 skip.
• Full workspace sweep: pytest tests/ -q -> 3010 pass, 26 skip, 0 fail.
• ruff check . clean (repo-wide, not just touched files).
• python scripts/validate_sync.py --verbose -> TOML schema valid, 66 controls, 7 handlers in sync.

Relationship to #406

Complementary:

• fix(baseline): enumerate real candidate paths in llm_eval passes #406 (fix-402-llm-eval-file-contents): fixes what the LLM tier sees WHEN it fires. TOML-only inside llm_eval blocks.
• This PR (widen-pattern-file-lists): fixes how often the LLM tier fires. TOML-only inside pattern blocks.

Either can merge independently. Together they close the tqdm-style failure mode from both sides: the pattern tier is more likely to resolve conclusively, and when it doesn't the LLM tier now sees real content.

Not in scope

• Framework code changes to the pattern handler's file resolution.
• Auto-derivation of .rst/.txt variants from .md entries in the pattern handler.

Both are separate follow-ups.