Keep Jev auditor requests within input limit

[quolu]

Cause

On mac-kite, claude-spotter@1.8.0 sent all 306 Codex catalog entries to Jev in one request. The API returned HTTP 400 with error_type=max_tokens_exceeded. Codex hook errors were projected as E_SPOTTER_AUDIT_GENERIC and recorded under SPOTTER.AUDITOR.UNAVAILABLE.

Change

• Pack Jev questions into requests of at most 32 KiB, preserving candidate order and the existing question format.
• Apply one timeout to the full audit and combine only complete, validated results and usage. Any failed batch fails the whole audit without a backend fallback.
• Document the contract and add tests for large catalogs and a later batch failure.

Verification

• node --test test/jev-backend.test.mjs: 12 passed.
• npm test: 607 tests, 604 passed, 3 platform skips, 0 failed.
• npm run verify:docs: passed.
• Mac installed 1.8.0, actual dotagents Codex catalog (306 entries): before change HTTP 400 max_tokens_exceeded; after change nine requests, all 306 covered, auditor CLI returned pass=true in 1.8 seconds. Runtime error sequence stayed at 228 after the smoke test.
• Manual full CI passed on macOS native, Linux workstation, and Windows native. PR Keep Jev auditor requests within input limit #4 is stacked on the pending CI contract PR Use workstation Linux only for product full CI #3 so its full matrix excludes linux-server. The initial main-based run was cancelled.
• The source file was hotpatched into the installed 1.8.0 package on mac-kite, rabbit, fox, and main-server. Each installed file has SHA-256 4b9811184400754fe6f6389743ca7abb6e69b3dd8dfcd7136984be0d74adbc98 and a dated backup. A package reinstall will replace the hotpatch until a versioned release includes this PR.