ci: Build the release's AppImage with Rust libobs on a manual dispatch too

[zackees]

Today's manually dispatched release (run 38497010624) failed every AppImage distro test: expected 'OBS-Studio-Rust 0.1.0-beta.20261014 (based on OBS ...)' from --version, got: OBS Studio - 33.0.0-beta6-1045-ga7edb3bc9. The AppImage was built without ENABLE_RUST_LIBOBS. Inside a called workflow github.event_name is the caller's event, so github.event_name != 'workflow_dispatch' || inputs.rust_libobs read a dispatch-only input that a workflow_call never sets. A scheduled release would have been fine.

• RED: ci/test_appimage_call_inputs.py checks that workflow_call declares rust_libobs (default true) and that RUST_LIBOBS doesn't depend on github.event_name.
• GREEN: rust_libobs is added to workflow_call, and RUST_LIBOBS: ${{ inputs.rust_libobs != false }} treats an unset input (push, tag, pull_request) as on.

[zackees]

The merged expression regresses direct push/PR/tag-style builds: GitHub's official @actions/expressions evaluator 0.3.61 evaluates inputs.rust_libobs != false as false when the inputs dictionary has no rust_libobs property. This matches GitHub's documented loose numeric equality (null/empty string and false all coerce to 0). The new test checks only input presence and absence of an event-name check; it does not evaluate the absent-input result.

I reproduced this with the actual source expression for push, pull_request, workflow_dispatch and schedule contexts: absent=false, explicit false=false, explicit true=true in each case. Absent input should preserve the ordinary Rust build default. The #664 AppImage-input work already owns this unit; I will integrate an actual evaluator regression and a fix that distinguishes absent input from explicit false there, avoiding a duplicate PR.

Sources: https://docs.github.com/en/actions/reference/workflows-and-actions/expressions and https://github.com/actions/languageservices/tree/main/expressions.