bug: Count binary patches in review limits

[idy]

Closes #49

Summary

• Use a binary-aware Git patch for the shared reviewer's 5,000,000-byte automatic gate, 100,000,000-byte hard cap, and effective-diff identity.
• Keep binary patch bodies out of model-facing per-file review chunks.
• Add a regression fixture where a binary model addition has a tiny plain diff but exceeds the automatic limit.

GizClaw/h6infra PR #4 demonstrates the bug: its plain Git patch is 4,147,843 bytes, while git diff --binary is 18,621,002 bytes. With this change it requires the existing exact-head admin approval command.

Validation

• node .github/scripts/pr-review/test.mjs
• node .github/scripts/review-request/test.mjs
• node .github/scripts/pr-readiness/test.mjs
• node .github/scripts/issue-review/test.mjs
• actionlint -ignore 'property "workflow_(repository|sha)" is not defined' .github/workflows/*.yml
• git diff --check

[github-actions]

✅ OpenAI PR Review: PASS

Conclusion: Ready from the OpenAI review perspective. PR format, linked Issue design, and code/plan conformance passed with no actionable findings.

Review checks

Check	Result
PR format	✅ PASS
Issue design	✅ PASS
Code & plan conformance	✅ PASS

Scope: f3f2e03017 · d371c53225..f3f2e03017 · full · 1 diff chunk

Usage: 53s · 270,653 tokens · 78.6% cache hit · credits unavailable

Summary

The complete diff has one reviewed chunk. It conforms to Issue #49’s plan, with no actionable code findings or plan-conformance blockers.

Review metadata

• Commit: f3f2e03017
• Range: d371c53225..f3f2e03017
• Mode: full
• Diff chunks: 1
• Model: gpt-6-sol
• Reasoning effort: low
• Session: repo-1309321116-pr-50-v2
• Generation: 9ab4655615d68316e35a8c30a4b98876d2d4ff9bd311477261a61761cbc886a0
• Evidence: 74947b5417ee19b90ae4662835a8bf31352a95343b8034e3b253c66a66c285d0

Totals

• Input: 268,153
• Cached input: 210,792
• Cache write: 56,545
• Output: 2,500
• Reasoning: 342
• Total: 270,653
• Estimated credits: unavailable for this model

Token and cache details

Stage	Mode	Target	Time	Input	Cached	Hit	Output	Total	Credits
pr	deterministic	format rules	0s	0	0	0.0%	0	0	N/A
pr	full	pr	11s	43,345	27,783	64.1%	474	43,819	N/A
issue	full	Issue #49	19s	78,523	70,494	89.8%	933	79,456	N/A
code	full	chunk 1/1	16s	81,824	51,588	63.0%	754	82,578	N/A
code	full	aggregate	7s	64,461	60,927	94.5%	339	64,800	N/A