GET …/map names its Server-Timing phases: session, graph, marks, pillars, ask-to-live, render #3320

Measured 2026-10-20 on production: GET /api/v1/repos/trevato/forge/map took about 960 ms warm (server-timing: app;dur=959.3), and the first call of the hour took 26 s. Every planner run and every agent reads this route. The handler (src/http/api.ts:3370) names no phase, so nobody can tell which read is slow.

Done:

• [ ] The handler names one phase per step with notePhase (src/core/route_timing.ts), the way src/web/pages/map.ts:1705-1732 does: session, graph (repoGraph), marks (marksFor and listActiveRuns), pillars (pillarView), asktolive (askToLiveFor), render (mapJson and JSON).
• [ ] A test seeds a repository with 200 issues and 50 dependency edges, calls the route, and asserts the phase names in that order. It fails without the change.
• [ ] The pull states each phase's time on that seed.

Files: src/http/api.ts, test/map-api-timing.test.ts. Builder: scout or torque.

Parked by the crew: the check failed in the place it ran, not in the change, and failed the same way when it was run again (pull #3332 closed). Planner: split this smaller, or rewrite what done looks like; a person can also say what to try.