fix: require transcription readiness before recording

[nikolaykolibarov]

Summary

• block recording before audio capture when the selected local model is missing or OpenAI is not configured
• prevent settings/onboarding from persisting an undownloaded local model as active
• make local onboarding require a model download by removing the skip path
• improve microphone/accessibility onboarding for already-granted and denied states
• add support logs for transcription and clipboard-copy failures
• stop tracking generated build/.DS_Store
• bundle whisper-cli/local transcription runtime in release builds for macOS, Windows, and Linux

Why

Users reported that copy/paste did not work from the beta DMG. Logs showed recording reached audio save and then entered error before any clipboard/paste path ran. A later tester log showed the hotkey worked, but recording was blocked because the CI DMG did not include whisper-cli. Local transcription must be self-contained in release artifacts; models remain downloaded separately into the user data directory.

Testing

• go test ./...
• npm run build
• ruby -e "require 'yaml'; YAML.load_file('.github/workflows/release.yml'); puts 'yaml ok'"

Test Release

A new CI-built prerelease will be published as v0.4.1-beta.3 with bundled local transcription runtime.