Don't let reporting a failed parallel worker abort the rest of the run

[nohwnd]

Two flaky tests, one of which turned out to be a real bug in Run.Parallel.

Reporting a failed worker aborted the run

Invoke-InRunspacePool reports a worker that threw with Write-Error and keeps going, because the remaining files still have results worth reporting. Write-Error obeys the caller's $ErrorActionPreference though, so a caller running with Stop gets a terminating error instead, the loop aborts on the first failing file, and the results of everything that already finished are dropped. CI scripts commonly run with Stop.

$ErrorActionPreference = 'Stop'
Invoke-InRunspacePool -InputObject @(1,2,3) -ThrottleLimit 3 -ScriptBlock {
    param($item) if (2 -eq $item) { throw 'worker blew up' }; $item
}
# before: THREW, nothing returned
# after : 1,3 returned and the error reported

Both Write-Error calls are now explicitly -ErrorAction Continue, with a test that sets Stop itself so it stays fixed regardless of the harness.

Pester's own test.ps1 sets Stop on line 43, so this reliably killed the P phase locally and with it the entire RSpec phase after it. With the fix the full local suite completes: 2915 passed, 0 failed, 3 skipped.

Timing assertions

Should-BeFasterThan.Does not throw when actual is faster than expected failed on PS7 - macOS latest on both #2993 and #2995 today and needed manual reruns.

Both flaky assertions bound the measured time from above, and measuring a scriptblock times compiling it, GC, and whatever else the machine is doing, which has no upper bound on a shared runner. Widened both from 100ms/1000ms to 30s, with a comment recording the asymmetry so they do not get tightened again: a 10ms sleep that takes 30 seconds means the machine is broken, and the opposite direction needs no margin because a sleep never finishes early.

🤖