Summary

• Skip redundant cache uploads when the restore step already matched the primary key, avoiding unnecessary work and API calls on every run.
• Track the matched restore key via CACHE_MATCHED_KEY_STATE and compare it against the primary key before saving.
• Add unit tests covering restore hit/miss, save skip, and error paths.

Approach

This follows the same pattern used by the official GitHub Actions that wrap @actions/cache:

• actions/cache
• actions/setup-python
• actions/setup-node
• actions/setup-go

All of these save the restored key via core.saveState() during the restore step, then compare it against the primary key in the post-run save step. If they match, the cache already exists and saving is skipped.

cacheId !== -1 check and removal of the "already exists" branch

The previous code caught errors from cache.saveCache() and checked message.includes('already exists') to handle duplicate saves. This branch was dead code in two ways:

1. @actions/cache catches ReserveCacheError internally (both v1 and v2 code paths) and logs via core.info() before returning -1. The error never propagates to the caller.
2. The ReserveCacheError message is "Unable to reserve cache with key ..., another job may be creating this cache.", which does not contain the string "already exists".

The new code checks cacheId !== -1 to align with the library's actual contract. Since @actions/cache already logs the reason internally on every path that returns -1, no additional logging is needed from our side. The catch block now only handles genuinely unexpected errors.