Windows runner randomly dies in `actions/cache`

I'm running into an issue where my Windows runners start randomly dying in action/cache. There doesn't seem to be any info or logging here, just dies with no messages and exits the build. This is using the standard windows-latest as of May 8 2026.

The workaround is setting continue-on-error for the actions/cache step, but it's a very strange failure.

Run actions/cache@v5
  with:
    path: ~/.cargo/registry/index/
  ~/.cargo/registry/cache/
  ~/.cargo/git/db/
  
  
    key: 2-Windows-cargo-nightly-e97b13b011c1ac01445c1af4f0f08dbde7bb36c1054a4f00050f6b2978bc56f8
    restore-keys: 2-Windows-cargo-nightly-
  
    enableCrossOsArchive: false
    fail-on-cache-miss: false
    lookup-only: false
    save-always: false
  env:
    CARGO_TERM_COLOR: always
    CACHE_KEY: 2
    CARGO_INCREMENTAL: 0
    CARGO_PROFILE_DEV_DEBUG: 0
    RUST_BACKTRACE: short
    RUSTFLAGS: -D warnings
    CARGO_UNSTABLE_SPARSE_REGISTRY: true
    CARGO_REGISTRIES_CRATES_IO_PROTOCOL: sparse
    CACHE_ON_FAILURE: true
Cache hit for: 2-Windows-cargo-nightly-e97b13b011c1ac01445c1af4f0f08dbde7bb36c1054a4f00050f6b2978bc56f8

The raw logs don't appear to show anything more interesting, just confirming that about one second after the cache starts restoring, the job exits:

2026-05-08T21:53:54.3339230Z env:
2026-05-08T21:53:54.3339485Z   CARGO_TERM_COLOR: always
2026-05-08T21:53:54.3339802Z   CACHE_KEY: 2
2026-05-08T21:53:54.3340077Z   CARGO_INCREMENTAL: 0
2026-05-08T21:53:54.3340397Z   CARGO_PROFILE_DEV_DEBUG: 0
2026-05-08T21:53:54.3340736Z   RUST_BACKTRACE: short
2026-05-08T21:53:54.3341057Z   RUSTFLAGS: -D warnings
2026-05-08T21:53:54.3341443Z   CARGO_UNSTABLE_SPARSE_REGISTRY: true
2026-05-08T21:53:54.3341900Z   CARGO_REGISTRIES_CRATES_IO_PROTOCOL: sparse
2026-05-08T21:53:54.3342337Z   CACHE_ON_FAILURE: true
2026-05-08T21:53:54.3342643Z ##[endgroup]
2026-05-08T21:53:54.7638812Z Cache hit for: 2-Windows-cargo-nightly-e97b13b011c1ac01445c1af4f0f08dbde7bb36c1054a4f00050f6b2978bc56f8
2026-05-08T21:53:55.3402022Z Post job cleanup.
2026-05-08T21:53:55.3590617Z Post job cleanup.
2026-05-08T21:53:55.7159737Z Cache up-to-date.
2026-05-08T21:53:55.7164287Z (node:1288) [DEP0040] DeprecationWarning: The `punycode` module is deprecated. Please use a userland alternative instead.
2026-05-08T21:53:55.7165604Z (Use `node --trace-deprecation ...` to show where the warning was created)
2026-05-08T21:53:55.8635363Z Post job cleanup.

https://github.com/mmastrac/linktime/actions/runs/25581305152/job/75100656575

Start by reproducing the linked workflow using actions/cache@v5 on a windows-latest runner, focusing on the restore and post-job cleanup stages shown in the logs. Compare runs with and without continue-on-error and determine why the job exits without an error; done means the failure is reproducible or explained and no longer causes a silent job termination.