fix(installer): stop PS 5.1 stderr wrapping from aborting version probes

[MasterFede5]

What does this PR do?

Fixes a Windows PowerShell 5.1 failure mode in install.ps1: the script sets $ErrorActionPreference = "Stop" globally, and PS 5.1 (the default powershell.exe the README instructs users to run) wraps redirected native stderr in ErrorRecords. The two --version 2>&1 probes (candidate at ~L265, installed binary at ~L288) therefore abort the whole install for any healthy binary that prints a single warning line to stderr while exiting 0 — reporting error: downloaded binary failed to run.

The fix relaxes EAP to Continue only around the two native probe calls, restoring it in a finally block. Failure detection is unchanged: it stays on $LASTEXITCODE.

Reproduction (PS 5.1.26100.8875, powershell.exe -File): stub binary printing a version to stdout + one warning to stderr, exit 0 → previous pattern aborts with the stderr text as the exception; new pattern captures the version output and still detects a nonzero exit (tested with exit 3). EAP is restored to Stop afterwards.

The literal invariants checked by tests/test_windows_bundle_contract.sh (& $DownloadedLauncher --version, launcher-only invocation) are preserved — that contract test passes locally.

Checklist

• Every commit is signed off (git commit -s) — required, CI rejects
  unsigned commits (DCO, see CONTRIBUTING.md)
• Tests pass locally (make -f Makefile.cbm test) — not run: no C toolchain on this machine; the change touches only install.ps1. tests/test_windows_bundle_contract.sh (the suite that asserts installer invariants) passes locally.
• Lint passes (make -f Makefile.cbm lint-ci) — C linters not applicable to this script-only change; install.ps1 parses clean (PSParser, 0 errors).
• New behavior is covered by a test (reproduce-first for bug fixes) — reproduced first with a stderr-writing stub as described above; the existing bundle contract test still guards the probe invariants. Happy to add a harness test if you point me at the preferred place for installer regression tests.

🤖 Generated with Claude Code

[DeusData]

Thanks for this, and sorry for the delayed acknowledgement. Queued for review.

A +19/-2 single-file change is easy to review — nothing blocking from a size standpoint. CI shows 1-2 failing checks worth checking while it waits.

PowerShell 5.1 wrapping stderr and thereby aborting verification is a nasty class of bug: it makes a working install look broken for reasons that have nothing to do with the install. Good catch.