feat: transcription dictionary learned from log edits

[rsslldnphy]

Summary

• Adds a WordDictionaryStore that persists a personal list of words to UserDefaults (com.wisp.wordDictionary)
• Transcription Log entries are now editable inline (double-click) — words present in the edited text but not the original are extracted and added to the dictionary automatically
• Dictionary words are injected into WhisperKit's promptTokens and the LLM cleanup prompt on every transcription, biasing the model toward the user's preferred spellings
• New "Transcription Dictionary" section in Preferences lets users view, add, edit, and delete words manually

Test plan

• Double-click a log entry, correct a mistranscribed word, press Save — word appears in Preferences → Transcription Dictionary
• Blank/whitespace edit does not add anything to the dictionary
• Add a word manually in Preferences → Dictionary, quit and relaunch — word persists
• Delete a word in Preferences → Dictionary — word is gone after relaunch
• swift test --filter "WordDictionaryStoreTests|WordExtractionTests|LogEntryEditTests" — all 44 tests pass

🤖 Generated with Claude Code