Address a recurring set of flaky unit tests. Two failure modes:
- Wall-clock perf microbenchmarks that assert absolute per-call timings.
These carry no correctness signal and flake on shared CI runners under
load. Removed or replaced with behavioural assertions.
- Testcontainer-backed tests whose cold container startup on a loaded
runner intermittently exceeded the per-test timeout. Raised the
timeouts (several were at 60s; one had no override and ran on vitest's
5s default).
Chris Arderne
264b1506a7