fix(contextful-gate): give each part a 2700 s exec deadline

[debuggingfuture]

Each contextful gate part now gets a 2700 s exec deadline instead of 1800 s.

Problem & Insight

• features.binary-*, workspace.cli, evaluate and budget.full each need 1250–1750 s on a good run: a cold build of 550–760 s (a release build of 1145–1518 s for budget.full) plus the suite. The ones above 1800 s end in ExecTimeout and report no verdict (contextful#465).
• Splitting those parts into more parts makes the whole gate slower. Parts run two at a time (STAGE_CONCURRENCY = 2), and each extra part repeats its own cold build.

Take

• STAGE_TIMEOUT_SEC = 2700 in runs/contextful-gate.ts. The slowest passing part so far ran ~1750 s, about 65% of the new deadline.
• check.limits.maxDurationSec goes from 3600 to 4500, keeping the same 1800 s of checkout, upload and retry headroom above a full exec.
• The batch wait is derived as queue age + the check run ceiling (now 165 min), replacing the hard-coded 130 min, so a part still inside its own budget is never reported failed.
• A test pins the 2700 s deadline each part receives, under the check run's ceiling.

Discovery execs keep 1800 s; they run in minutes.

[flaredispatch-fractalboxdev]

⚠️ pr-review could not complete: { "step": "review", "cause": Error: ModelCallFailed: openai returned 429: {"success":false,"result":[],"messages":[],"error":[{"code":2003,"message":"Rate limited"}],"name":"AiGatewayError","httpCode":429,"internalCode":2003,"message":"Rate limited","description":"Rate limited"}, "summaryMd": undefined }

📋 View full logs & reviewed diff ↗