Full audit and rewrite of extension — 22 bugs fixed #4023

Originally written by @ccantynz-alt on GitHub.
Imported from https://github.com/ccantynz-alt/48co.v1/pull/14

OFFSCREEN.JS:

• Pre-flight API key validation BEFORE recording starts
• Errors sent as { error: '...' } not as transcript text
• Audio streams properly released on every failure path
• Specific error messages (401=bad key, 429=rate limit, timeout)

BACKGROUND.JS:

• Tracks which tab STARTED recording (recordingTabId)
• Transcription results sent to recording tab, not active tab
• ensureOffscreen() wrapped in try/catch — reports failure to content
• New OFFSCREEN_STARTED acknowledgment message
• New GET_RECORDING_STATE for popup to check on open
• Error badge state added

CONTENT.JS:

• Separate timeout IDs for recording (60s) vs processing (30s)
• State guard: processTranscript only runs if in recording/processing state
• State guard: recognition.onend checks state before processing
• TRANSCRIPTION_READY ignored if not in expected state
• resetToIdle() helper clears ALL timers and recognition
• Web Speech errors show error badge, not silent console.warn
• insertText: 3 methods — execCommand → clipboard paste → DOM insert
• START_RECORDING checks response for errors
• Empty transcript shows error badge briefly instead of silent reset

POPUP.JS:

• No optimistic UI updates — waits for real state from badge
• Checks GET_RECORDING_STATE on open
• Shows "click to force-reset" when stuck on Transcribing
• Error state shown with red indicator and retry prompt
• forceReset() sends to active tab content script

https://claude.ai/code/session_01PBeNmPHrD2W6A2YKZRC7V3