Collect a hang dump on the wedging Windows PowerShell CI leg

[andyleejordan]

What

CI's windows-latest leg still intermittently wedges and rides its
timeout-minutes: 60 cap to a cancelled CI Tests run, even after #2318's
discovery-time skips. This PR doesn't add another skip — it instruments CI so
the next hang is actionable instead of burning an hour blind.

Where it wedges now

Pulling the Windows test-result artifacts from the last two hung runs and a
passing one (TestFull order is TestPS74 → TestE2EPwsh → TestPS51 → …):

Run	Result	Legs that produced a .trx	net462 TestPS51?
#2319 e34a3ffee	✅ ~11 min	7 (incl. net462 TestPS51, 338 tests)	yes
#2325 dbd5dfc	❌ 1 h cap	2 (TestPS74, TestE2EPwsh)	no
#2319 7e0aa34	❌ 1 h cap	2 (TestPS74, TestE2EPwsh)	no

Both hung runs stop right after TestE2EPwsh and never emit a TestPS51
.trx, so they wedge in the net462 / Windows PowerShell 5.1 unit leg —
not the E2E server path #2318 skipped, and not covered by any
SkippableFactOnWindowsPowerShell guard (those all live in the E2E project).
All three commits contain #2318's skip (verified via merge-base), so this is
not a missing rebase — the 20260614 runner-image regression still races,
just less often. See #2323 for the image root cause.

Changes

• PowerShellEditorServices.build.ps1 — add
  --blame-hang --blame-hang-timeout 10m --blame-hang-dump-type full to the CI
  dotnet test invocations, gated on $env:GITHUB_ACTIONS (local runs are
  byte-identical). Any single test that wedges past 10 minutes is dumped and
  its host process tree terminated, so the leg fails fast and names the
  test instead of riding the cap.
• .github/workflows/ci-test.yml — best-effort ProcDump install on the
  Windows leg (VSTest's built-in hang dumper only handles .NET Core hosts, so
  capturing a dump of the net462 host needs ProcDump, which isn't on the runner
  image; a download failure only warns and never fails CI), and extend the
  test-results upload to **/*.dmp + **/*_Sequence.xml.

Caveat

If the wedge is in host startup/discovery rather than a running test, the
per-test timer may not fire — but _Sequence.xml will still show how far
discovery got, which is itself diagnostic. Once a hung run identifies the
offending net462 unit test, we can give it the same targeted
Skip.If(Desktop) / polling fix we used for #2307 and #2314 instead of a
broad leg-wide skip.

Validation

Local arg construction verified in both modes (CI inserts the flags with
--framework still last; local is unchanged); build script pwsh-parsed and
workflow YAML validated. The instrumentation itself can only be exercised on
the GitHub-hosted Windows image where the hang reproduces.

_{Drafted by Copilot (Claude Opus 4.8) on Andy's behalf.}