Get word timestamps from Apple transcript JSON exports

Apple's transcript JSON (from Voice Memos/Notes recordings) has no plain segment list. Text and timing sit in attributedString: runs alternates a text fragment with an integer index into attributedString.attributeTable, and each table entry holds timeRange as [start_seconds, duration]. Walk runs in pairs, keep each fragment's start time and character offset, then join the fragments into one string. Searching that string for a phrase and mapping the match offset back to the nearest start time gives mm:ss for any passage. That's useful for pointing a user at the exact audio to re-listen to where auto-transcription garbled words.