PR Summary

Fixes #28375.

The module analysis cache is written by a background task that waits at least 13 seconds after the last change, so a pwsh that runs one command and exits sooner (pwsh -Command, pwsh -File) never saves it, and every such process analyzes the same modules again when a command is auto-loaded.

When QueueSerialization queues the first save, it now also registers a ProcessExit handler that writes the cache if the save is still pending. A lock keeps the handler and the delayed write from running at the same time. The delayed write is unchanged for long-running sessions. Once the cache is up to date nothing is queued, so later processes do not write it again.

PR Context

Measured on Windows 11 26100 (i9-10900K), Release builds of master 1aac7f5 with and without this change (the two differ only in System.Management.Automation.dll), default cache location, 40 interleaved runs, median of paired differences with 95% confidence interval:

The first run writes the cache (4 KB here); later runs did not rewrite it.

Tests: added test/powershell/engine/Module/ModuleAnalysisCache.Tests.ps1, which starts pwsh with PSModuleAnalysisCachePath in TestDrive, auto-loads Get-ChildItem and checks that the cache file exists after the process exits. It fails without the change and passes with it. CommandDiscovery, Import-Module, ModulePath, Base-Directory and TabCompletion tests give the same results with and without the change on my machine.

I used GitHub Copilot to investigate this and to help write the change and test; I reviewed the code and ran the tests above myself.

PR Checklist

• PR has a meaningful title
• Summarized changes
• Make sure all .h, .cpp, .cs, .ps1 and .psm1 files have the correct copyright header
• This PR is ready to merge.
• Breaking changes
  • None
• User-facing changes
  • Not Applicable
• Testing - New and feature
  • Make sure you've added a new test if existing tests do not effectively test the code changed