fix(test): give farm.test.ts a file-level budget instead of chasing blocks (#542)

[SUaDtL]

Third pass on #542, and the last — because the budget finally matches the population it applies to.

What kept happening

Every case in farm.test.ts launches a real farm subprocess. Process creation on a hosted Windows runner is far slower than on a developer box, so the 5000 ms default is marginal for the whole file — not for any one case in it.

pass	fix	what happened next
1	raised the single case #542 named (3553 ms isolated, 5331 ms under CI load)	seven more in its block timed out
2	raised that block	two cases in a different block timed out
3	this one — the file	—

Each pass fixed the sample the last failure happened to name, and the next-slowest case was always waiting behind it. That is the actual lesson, and it cost two round trips through CI to learn: when a failure names one member of a population, size the fix to the population.

The fix

vi.setConfig({ testTimeout: 30_000 }) for this file. ~6× the slowest measured run and still bounded, so a genuine hang fails rather than sitting until the job timeout.

Scoped to this file rather than vitest.config.ts deliberately — the sibling suites are ordinary in-process unit tests, and a 30 s default there would turn a hung unit test from a 5 s failure into a 30 s one.

The block-level timeout from pass 2 is removed: two sources of truth for one budget is how they drift apart.

Verification

537 tests pass across all 9 files. Test-only — the payload gate confirms no shipped change, so no version bump.

The real proof is this PR's Coverage union <os: windows-latest> cell, which is what caught both previous incomplete fixes.