Explore answers under 50 ms and says so in Server-Timing, measured by a test #10946

Intent: /explore, /api/v1/explore and the MCP explore tool answer from one ranked read that is fast and measured.

Done:

• [x] A test seeds 200 repositories and asserts the ranked read under a stated budget, failing when the budget is exceeded.
• [x] /explore carries a Server-Timing header with the read phase.
• [x] If the read is slower than budget, cache the ranking per viewer-visibility for a few seconds as docs/CACHING.md describes, with the cache key naming what the viewer may see (a private app never leaks through it).
• [x] A test proves a private app is not shown to another viewer from the cache.
• [x] Budget registered where src/perf/budgets.ts keeps the others.

Files: src/actions/explore.ts, src/perf/budgets.ts, test/explore-budget.test.ts. Risks: a cache keyed too coarsely leaks private rows; measure first, cache only if over budget. Brief: torque (performance).

Reviewed pull #10955 at its head: ticked A test seeds 200 repositories and asserts the ranked read under a stated budget, failing when the budget is exceeded.; /explore carries a Server-Timing header with the read phase.; If the read is slower than budget, cache the ranking per viewer-visibility for a few seconds as docs/CACHING.md describes, with the cache key naming what the viewer may see (a private app never leaks through it).; A test proves a private app is not shown to another viewer from the cache.; Budget registered where src/perf/budgets.ts keeps the others..