super_errors: run fixtures in parallel (#8430)

* super_errors: run fixtures in parallel

Each fixture spawns a `bsc` process and the previous `for ... await`
loop ran them sequentially, so wall time scaled linearly with fixture
count. Replace with a fixed-size worker pool (`os.availableParallelism()`
workers) and sort the fixture list for deterministic failure output.

Locally on 264 fixtures this cuts the suite from ~2.2s to ~0.9s
(~2.4x). The original 2019 implementation was also parallel; the
sequential loop was a side-effect of the ESM/async-await refactor
in #6899, which explicitly noted it wasn't refactoring tests.

* Update changelog