Make a parallel run print what a sequential run prints

[nohwnd]

Fix #3044

Two things made Run.Parallel print something different from a sequential run of the same files.

Every warning was printed twice, and three times for the workers that started first. Invoke-InRunspacePool creates the runspace pool with the caller's $Host, so a worker's Write-Warning already reaches the console while the worker runs. The loop that re-emitted Streams.Warning after EndInvoke printed it again. Errors keep their re-emit, they only land in Streams.Error and never reach the host on their own. Regression from pester/Pester#2987 - Run test files in parallel on Windows PowerShell 5.1 too.

A worker ran with default preference variables. A fresh runspace has WarningPreference = 'Continue' and VerbosePreference = 'SilentlyContinue' no matter what the caller was running with, so a caller who had silenced warnings still got them and a caller who had asked for verbose output got none. Invoke-Pester now hands its caller's session state down, and the worker runspaces are defined with WarningPreference, VerbosePreference, DebugPreference, InformationPreference and ProgressPreference taken from it. This one is not new in 6.2.0, 6.1.0 lost the verbose output the same way, warnings only looked right there because ForEach-Object -Parallel merged the worker's records into the parent's stream where the caller's preference filtered them.

One test writing to every stream, counting how many times each line reaches the console:

Stream	serial	parallel 6.2.0	parallel, this PR
Write-Host	1	1	1
Write-Error	1	1	1
Write-Warning	1	2	1

and with the caller setting WarningPreference = 'SilentlyContinue', VerbosePreference = 'Continue':

Stream	serial	parallel 6.2.0	parallel, this PR
Write-Warning	0	1	0
Write-Verbose	1	0	1

Write-Host is unaffected throughout, it is not preference-controlled and already printed once in both modes.

Added tests for both. The warning one fails with the re-emit loop put back, the preference one fails with the wiring removed. The preference test runs in a child process, because a worker writes to the shared host and no redirection inside the test process can see that.

• ErrorActionPreference is left out on purpose, it decides how a command behaves on error and not only what reaches the screen. A sequential test body does inherit it, so this is a known remaining difference. Worth folding in, or leaving?

🤖

[kborowinski]

@nohwnd So the workers will surface warnings anyway but only when running tests in parallel? I went through the changelog for 6.2.0 and didn't see that. Tests running serially or in parallel on 6.0.0, 6.1.0 do not surface warnings. Was this a bug, that the warnings were not surfacing or the current change is a breaking change?

Tests running serially and then parallelly on Pester 6.2.0:

Warning output surfacing only when running in parallel

vs tests running serially and parallelly on Pester 6.1.0:

Warning output not surfacing at all