Windows CI: vm-config lib test process crashes (STATUS_ACCESS_VIOLATION 0xc0000005)

[mudcube]

Context

After CI began running the suite (the working-directory: rust fix in #63), the windows-latest leg surfaced a hard crash unrelated to that PR's changes. The Windows CI leg is currently advisory (continue-on-error) so it doesn't block merges; this issue tracks diagnosing/fixing the crash.

Symptom

cargo test -p vm-config --lib exits abnormally with a memory access violation — not a normal assertion failure:

error: test failed, to rerun pass `-p vm-config --lib`
Caused by:
  process didn't exit successfully: `...\vm_config-*.exe --test-threads=10`
  (exit code: 0xc0000005, STATUS_ACCESS_VIOLATION)
note: test exited abnormally; to see the full output pass --no-capture to the harness.

Several detector::tests::* pass first, then the process crashes. Because the run uses --test-threads=10, the exact offending test isn't pinned.

Investigation notes

• Re-run on Windows with --test-threads=1 --no-capture to isolate the crashing test.
• A 0xc0000005 in safe Rust usually means a stack overflow (deep recursion) or a bug in a native/unsafe dependency. This PR (Update all dependencies, fix clippy/tests, and repair CI #63) bumped many deps (git2 0.21, nix 0.31, sysinfo 0.39, etc.) — a Windows-specific regression in one of them is plausible. cargo tree -p vm-config + bisecting deps is a likely path.
• Unrelated but adjacent: validate::tests::test_valid_config was also failing on Windows due to a Unix-only starts_with('/') absolute-path check — already fixed in Update all dependencies, fix clippy/tests, and repair CI #63 (now uses Path::is_absolute()). This crash is separate.

Evidence

Failing run (commit d67c747): https://github.com/goobits/vm/actions/runs/28057405127/job/83063072548

Done when

The vm-config test binary runs to completion on Windows and the continue-on-error advisory flag can be removed from the Windows matrix leg in .github/workflows/ci.yml.