fix(evidence): expose bounded reviewer timeouts

[scarmani]

Summary

• Add explicit --reviewer-timeout and --overall-timeout controls to review-queue collect-evidence and scripts/collect_quorum_evidence.py.
• Fail closed when the overall reviewer deadline expires, with JSON diagnostics for orchestration_timeout, timed_out_families, and overall_timeout_seconds.
• Apply per-reviewer timeout overrides to Claude, Codex/OpenAI, API reviewers, Grok Build, and Antigravity/Gemini CLI paths for only the current invocation.

Validation

• Current-main baseline: origin/main = 3d2af8d2c10f981d7ddd37cba6cc3f96af1afaba; python3 -m aragora.cli.main review-queue collect-evidence --help does not expose the timeout flags there.
• Exact PR head: 1bf591bdb41193bc9d12a1908583f3f97ce264ec; both python3 -m aragora.cli.main review-queue collect-evidence --help and python3 scripts/collect_quorum_evidence.py --help expose --reviewer-timeout and --overall-timeout.
• PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 timeout 300 /Users/armand/.pyenv/versions/3.11.11/bin/python -m pytest tests/swarm/test_quorum_evidence.py tests/cli/commands/test_review_queue.py -q - 476 passed.
• timeout 240 pre-commit run --files scripts/collect_quorum_evidence.py aragora/swarm/quorum_evidence.py aragora/cli/commands/review_queue.py aragora/cli/parser.py tests/swarm/test_quorum_evidence.py tests/cli/commands/test_review_queue.py - passed.
• After marking ready for review, required non-quorum checks are green (lint, typecheck, Generate & Validate, TypeScript SDK Type Check, sdk-parity). aragora-merge-quorum fails closed as expected because this is Tier 4 merge/evidence-authority code and still needs current-head model quorum plus human preapproval/settlement before merge.

Dogfood on blocked PR evidence flow

• feat(odr): in-package ODR verification engine (odr_verify) — ODR-3 server core #8389 merge-packet at 136f3002013c7026c74f68d1892c52718e33b23b: Tier 1, mergeable, required checks green except aragora-merge-quorum, blocker is missing model quorum/focused dogfood.
• Artificial fail-closed check on feat(odr): in-package ODR verification engine (odr_verify) — ODR-3 server core #8389 through both public entrypoints with --reviewer-timeout 1 --overall-timeout 0.01 --json: returned action=prepare, orchestration_timeout=true, timed_out_families=[claude,grok], overall_timeout_seconds=0.01, zero evidence items, and no posted families.
• Bounded real dry-run on feat(odr): in-package ODR verification engine (odr_verify) — ODR-3 server core #8389: python3 -m aragora.cli.main review-queue collect-evidence --repo synaptent/aragora --pr 8389 --reviewers claude grok --reviewer-timeout 45 --overall-timeout 60 --json returned one countable Grok PASS (would_count=true) while Claude exceeded the 60s orchestration deadline. The collector still failed closed with action=prepare, orchestration_timeout=true, timed_out_families=[claude], has_supportive_quorum=false, and no posts.
• feat(routing): decision-stakes router — route by tier, record rationale (#8233) #8393 merge-packet at 7aacade51697f601c4e595325940143505ded01d: PR is closed, and the packet correctly reports not_ready_for_settlement; no reviewer evidence was collected for the closed PR.
• [AGT-04] GitHubEventResolver — concrete event-to-claim resolution adapter #8519 and chore(deps): consolidate httpx and sdk dependency bumps #8713 packet checks still report exact-head missing quorum; both are Tier 3 and require human risk settlement after model evidence.

Dogfood note

• During the feat(odr): in-package ODR verification engine (odr_verify) — ODR-3 server core #8389 Grok dry-run, the reviewer CLI left two untracked feat(odr): in-package ODR verification engine (odr_verify) — ODR-3 server core #8389 target files in the disposable worktree. They were removed before validation and are not part of this PR diff; the timeout feature still failed closed correctly. This is a reviewer-harness isolation follow-up, not a timeout-control blocker.

[github-actions]

Aragora Code Review

Advisory-only review. No issues found.