Fix OCR compatibility status and numeric limits for 0.8.3

[xeonvs]

Outcome

Toolkit 0.8.3 feature implementation is merged. This PR closes the repository feature gate for #139, #140, #142, and #143; stable publication and issue/milestone closure remain owned by the protected release workflow and independent external readback.

Delivered

• Authoritative, recoverable failed OCR compatibility publication.
• Installed-OCR ownership for max-tools through sentinel 0, plus behavioral numeric boundary qualification.
• Complete bounded native/context/evidence OCR tool-call activity in the existing GitLab technical summary, without per-tool token attribution.
• macOS Python 3.12/3.14 jobs remain visible best-effort diagnostics; Linux, coverage, quality, security, dependency, package, and CodeQL gates remain required.

Evidence

• Reviewed head: 6520a53db6b4f9d907735456be43e0b17ec296af
• Reviewed/merge tree: 6ab22194ecb44329a7b89be918c4c5afb8f8daf7
• Verified squash merge: 74b3f46f72f1d6231c44cc39ca12f732a46da706
• Local: 1,275 tests + 324 subtests; 86.35% branch coverage; risk groups 84/82/86/87 against 80/80/85/85 floors.
• Hosted: all current-head Linux and macOS jobs, quality, security, dependency, build, and CodeQL checks passed; zero review threads.
• Live main ruleset readback retained ten required non-macOS contexts and removed only the two macOS contexts.

No real LLM/model/provider request was run or claimed. Deterministic installed OCR 1.10.0 no-model boundary evidence remains the exact qualification boundary.