feat(message): add thinking blocks with live streaming

[lovec741]

Summary

• render model reasoning as collapsible thinking blocks: "Thinking…" + spinner while active, "Thought for Xs" when done, muted gray markdown body, tap header to expand/collapse
• stream reasoning live by applying message.part.delta to reasoning parts (server sends field: "text" regardless of target part type, so the reducer routes deltas by target part type)
• auto-expand thinking toggle in settings (default off, persisted); toggle is view-only — deltas always accumulate, so flipping it mid-stream reveals the backlog instantly
• snap reasoning body to full text when the part completes instead of trailing the typewriter into the answer
• show existing backlog immediately when expanding mid-stream (no replay from zero)
• follow list auto-scroll while reasoning grows; per-part expand state survives list rebuilds (mirrors tool expansion pattern)
• add EN/zh l10n strings and tests for reducer, dispatcher, widget, and settings persistence

Verification

• dart format .
• flutter analyze --no-pub
• flutter test --no-pub (184 tests, 6 new)

AI disclosure

This change was co-written by GLM-5.3-Flash.
I read and reviewed all of the code, tried to match it to the codebase conventions, and tested it on Android.

Notes

Thanks for the project! I have been using it for a couple of months and it has been very useful!
I want to add a couple of the features that I have been missing. I am starting with this one and I am considering adding message undo next.