Issues with `env.LOCKFILE` when `package-manager` input is explicitly specified but there is no package manager cache

When package-manager is explicitly set (e.g. package-manager: yarn), the "Check lockfiles" step in action.yml sets PACKAGE_MANAGER but never sets LOCKFILE. The auto-detection branches all set both; the explicit branch sets only one.

This causes cache-dependency-path in the "Setup Node" step to evaluate to "./" (a bare directory), which actions/setup-node
glob expands recursively to every file in the workspace, spawning a potentially large number of concurrent NodeJS processes, which in turn corrupts V8's startup snapshot cache and produces fatal crashes. See https://github.com/actions/setup-node/issues/887

Steps to reproduce

1. Use a repo where the lockfile is not committed (e.g.yarn.lock is in .gitignore)
2. Set package-manager: yarn explicitly in the workflow
3. Run the action — the Setup Node step crashes

Expected behaviour

The cache-dependency-path should resolve to the lockfile path (e.g. ./yarn.lock),
identical to the existing auto-detection path.

If the lockfile is absent, caching should be skipped gracefully with a warning, rather than crashing.

Note well

This issue has been prepared with the help of Claude Code and ChatGPT Codex. I'm working on a fix with them and hope to land a PR soon, after some more testing.

Start in action.yml at the "Check lockfiles" and "Setup Node" steps, then reproduce with an explicitly specified package manager and no committed lockfile. Verify that cache-dependency-path resolves to the intended lockfile path when present and that caching is skipped with a warning when it is absent, without causing the Setup Node step to crash.