fix: Stop docs and test tasks unloading the caller's modules

[tablackburn]

Summary

Fixes #221 and #222, which share a root cause: both Build-PSBuildMarkdown and
Test-PSBuildPester treated Remove-Module -Name as an undo for an import. It is not —
it removes every loaded module with that name, including a copy the caller loaded and
the function never imported. Both sit on default task chains (Build → BuildHelp →
GenerateMarkdown, and Test → Pester), so a build silently emptied the session it
ran in. PowerShellOrg/PSDepend dropped the documentation tasks entirely rather than
live with it.

• Build-PSBuildMarkdown unloads a module the caller had loaded #221 — Build-PSBuildMarkdown now records what was loaded, removes only the
  instance it imported, and restores what it displaced. That covers the zero-export path
  too, where it warns and returns without generating anything — return inside try
  still runs the finally.
• Test-PSBuildPester unloads a module it never imported #222 — Test-PSBuildPester only removes what it actually imported. Its import is
  conditional on -ImportModule (default $false) while its removal was not, so on the
  default Test chain it unloaded a module it had never touched.
• Build-PSBuildMarkdown also drops -Global from its import.

Three constraints, each measured on this branch:

• Scoping the removal alone is not sufficient. Import-Module -Force independently
  evicts a same-path copy before the finally is reached — measured
  same-path: instances=1; original instance object still present? False. Hence
  capture-and-restore rather than only changing the removal.
• -Force must stay. Without it, importing an already-loaded module returns the
  cached PSModuleInfo: measured WITHOUT -Force: Get-Alpha versus
  WITH -Force: Get-Alpha, Get-Beta after a rebuild. $moduleInfo goes straight to
  New-MarkdownCommandHelp -ModuleInfo, so dropping it would silently document a stale
  command surface.
• -Global on the docs import is unnecessary. PlatyPS 1.0.3 resolves the module
  through the PSModuleInfo object, never by name. Generated markdown compared
  byte-for-byte before and after: Get-Widget.md, Set-Widget.md, and
  PSBuildTestFixture.md all identical=True. Note -Global is kept on the restore
  import — an import issued from inside the module without it reaches only
  PowerShellBuild's own session state, and the caller would still see nothing.

Test Plan

./build.ps1 -Task Test — 597 passed, 0 failed, 3 skipped.

Six new tests, driven by a new Invoke-PSBuildModuleEvictionProbe fixture helper that
preloads a module in a background job, invokes the command, and reports what is loaded
and callable on both sides of the call:

• Build-PSBuildMarkdown unloads a module the caller had loaded #221 caller's module loaded from the documented path survives, Get-Widget still callable
• Build-PSBuildMarkdown unloads a module the caller had loaded #221 zero-export path (FunctionsToExport = @()) generates nothing and evicts nothing
• Build-PSBuildMarkdown unloads a module the caller had loaded #221 a session that started clean ends clean
• Test-PSBuildPester unloads a module it never imported #222 -ImportModule not passed leaves the caller's module completely alone
• Test-PSBuildPester unloads a module it never imported #222 -ImportModule passed restores the caller's copy and removes its own (the
  caller's copy and the copy under test are at different paths, so the count
  distinguishes evicted / left behind / correct)
• Test-PSBuildPester unloads a module it never imported #222 a session that started clean ends clean

Red-before-green verified by reverting only the two source files: the four eviction tests
fail with Expected 1, but got 0; the two "nothing loaded" tests pass either way, which
is their job — they guard against over-restoring.

Breaking Changes

None. This is a behavior fix, but it changes what a build does to your session on
upgrade, so it has CHANGELOG.md and docs/migration-v0.8-to-v1.0.md entries.

One residual cost, documented there: the restored module is a fresh import, not literally
the caller's original instance, so a PSModuleInfo reference held across the call goes
stale. That is the price of keeping -Force, and much smaller than the command
disappearing.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

Test Results

    4 files  ± 0    893 suites  +4   3m 45s ⏱️ + 1m 18s
  602 tests + 6    599 ✅ + 6   3 💤 ±0  0 ❌ ±0 
2 391 runs  +24  2 319 ✅ +24  72 💤 ±0  0 ❌ ±0 

Results for commit c36cc44. ± Comparison against base commit 3ea9e82.