Hello! I've been looking for a way to save the cache even if an intermediate step fails (#79). (Context: on Turing.jl docs it takes 10 mins to instantiate the environment and 2 hours to build the docs, so having this would save a lot of time on PRs)

I understand that a 'nicer' solution to this would depend on upstream fixes to GitHub Actions, but this PR seems like a good enough way to do it. The gist is that as long as we know the cache paths and the key, we can manually call actions/cache/save to save the cache at the bottom of the workflow. (That's also the workflow that GitHub said to use.)

So if julia-actions/cache outputs these as variables, it becomes easy to write something like

jobs:
  test:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4

      - name: Load Julia packages from cache
        id: julia-cache
        uses: julia-actions/cache@v2
        with:
          cache-name: foo

      # do whatever you want here (that might fail)

      - name: Save Julia depot cache
        id: julia-cache-save
        if: always() && steps.julia-cache.outputs.cache-hit != 'true'
        uses: actions/cache/save@v4
        with: 
          path: |
            ${{ steps.julia-cache.outputs.cache-paths }}
          key: ${{ steps.julia-cache.outputs.cache-key }}

I've tested this using a playground repo and a workflow pointing to my fork, and it seems to work perfectly as far as I can tell. You can see the first time the action runs here (no cache hit, and an intermediate step fails, but the cache is still saved):

https://github.com/penelopeysm/Shaymin.jl/actions/runs/14022967403/job/39257320572

and a subsequent run where the cache is then correctly loaded:

https://github.com/penelopeysm/Shaymin.jl/actions/runs/14022985223/job/39257363646

If you think this approach is sensible, then I can add some documentation explaining this too.