Chat reply chips that speak as you

This release replaces the old inline follow-up chip tool with a post-response reply system that behaves more like a real conversation.

Shipped commits

• fix(chat): generate reply chips after assistant responses - Suggested replies now appear only after LifeJet finishes answering, are generated from the recent thread plus LifeJet context, and are stored on the assistant message for reloads and review.
• fix(chat): make suggested replies send immediately - Tapping a chip now sends that text as the user's next message, and chips fade away as soon as the conversation moves forward.
• fix(chat): retire assistant-authored follow-up chips - The chat model no longer has the old suggestFollowUps tool, so it cannot produce awkward chips that ask LifeJet how the user's own experiment went.

Why it matters

Reply chips should lower the effort of continuing, not make the user translate LifeJet's wording back into their own. The new system keeps chips in the user's voice, lets the model use memory-aware context when it helps, and avoids showing stale options after the user has already replied.