[Stretch] Support multi-audio conversation merge

**Why:** Users may share multiple audios from a conversation; merge into single context.

**What to build:** Allow users to upload multiple audio files in sequence. Concatenate transcriptions with speaker/timestamp markers. Pass merged text to reply generation. Store conversation history in IndexedDB.

**Acceptance criteria:**
- [ ] Share target accepts multiple audio files (or sequential uploads)
- [ ] Each audio transcribed separately
- [ ] Transcriptions merged with timestamps ("Audio 1: [text] → Audio 2: [text]")
- [ ] Merged text passed to generateReply
- [ ] Conversation history saved to IndexedDB
- [ ] UI shows all audios in thread
- [ ] Clear button to reset conversation

**Depends on:** #13

Start by reviewing dependency #13 and tracing the current share-target upload, transcription, and generateReply flow. Then inspect how conversation state can use IndexedDB. Done means multiple audios appear in one thread, each transcription is merged with timestamps and passed to generateReply, history persists, and the clear action resets the conversation.

Written by the indexing model from the issue text.