[kvlonge]

Why

The native TOML config reader carried about 700 lines of shims reproducing Python tomllib edge cases, and its errors had lost column locations and mentioned tomllib. This PR also carries the 0.156.0 release memory exception for the 3,000-model lineage guards, which missed #1086: the accepted lineage memory growth puts them at 130–137 MiB against a 128 MiB cap.

Changes

• Report line and column for every TOML syntax error; a leading byte order mark is now accepted (also by the Python-side lint and skills readers).
• Remove the namespace, large-integer and deep-nesting shims; integers beyond 64 bits and nesting past 64 levels get clear errors. TOML 1.1 syntax is still rejected.
• Contract test: documents tomllib accepts still load identically; failure baselines updated.
• Raise the four 3,000-model model-lineage fresh-process RSS caps from 128 to 160 MiB, marked temporary until after the 0.156.0 release.

Verification

• Config crate tests, 10,000-document comparison, and CLI compile/format test with a BOM config pass.
• Failure-corpus differential (348 projects) and make check-ci pass; fensu check passes after the test-case fix.