feat: return tri-state outcome from inject_text instead of bool

[goodroot]

Follow up to #224

[github-actions]

Continue PR Review

Automated review via Continue CLI run through opub.

Summary

The tri-state refactor looks mostly consistent and the updated tests cover the main behavior changes. I did not find any obvious security issues in the diff. The main remaining concern is backward compatibility for any callers that still treat TextInjector.inject_text() as boolean, plus a missing direct regression test around main._inject_text()/capture mode semantics.

Findings

• lib/src/text_injector.py: TextInjector.inject_text() now returns an InjectionOutcome enum instead of bool. This PR updates the known call sites in main.py and longform_controller.py, but if there are any other internal or downstream callers comparing the result with is True / is False or serializing it, they will silently misbehave because enum truthiness differs from strict booleans. Before merge, grep for all callers and either update them or document this as a breaking API change.
• Missing regression test for lib/main.py::_inject_text() capture mode and consumed outcome handling. The new behavior treats both INJECTED and CONSUMED as non-failures and also maps capture-subscriber mode to InjectionOutcome.INJECTED; that’s reasonable, but currently only AST-shape tests verify main.py. A behavioral unit test should assert:
  • capture subscriber bypasses TextInjector and returns InjectionOutcome.INJECTED
  • InjectionOutcome.CONSUMED does not trigger failure/error UI paths
  • InjectionOutcome.FAILED still does

Suggested follow-ups

• Add a repository-wide grep/test pass for all inject_text( callers to confirm no boolean-only consumers remain.
• Add direct behavior tests for hyprwhsprApp._inject_text() and _process_audio() using mocked injector outcomes rather than only AST inspections.