feat(v0.11): Reasoning Trace Display — surface thinking blocks from extended reasoning models

[vivekchand]

Problem

Reasoning/thinking blocks from Claude are already parsed and rendered, but the UX is poor: thinking blocks are expanded inline with no visual separation from tool calls, there's no token count, and there's no way to collapse them.

What's Already There

• api_transcript_events() (~line 10719) extracts type: 'thinking' blocks from JSONL transcripts (lines ~10762-10763)
• The transcript modal renders thinking events with CSS class .evt-item.type-thinking (line 2072) — left border #6b7280, icon 💭
• api_transcript_events() truncates thinking text to 2000 chars per block (line 10763)
• Sessions list shows a 🧠 badge when thinking: true is set on a call (~line 7012)
• thinking_calls counter is tracked and surfaced in session stats (~line 13051)

What's Missing

1. Collapsed by default — thinking blocks should render collapsed (summary line only) with a toggle to expand; right now they render fully expanded inline
2. Token count — the API returns thinking token usage but it's not shown per-block or in aggregate in the transcript viewer
3. Visual grouping — when multiple thinking blocks appear in sequence (e.g. extended thinking), they should be grouped rather than listed individually

Where to Change

• Rendering: viewTranscript() (~line 5072) builds the message HTML — add a collapse/expand toggle for type === 'thinking' events
• Token display: api_transcript_events() can pass token counts if present in the raw JSONL block; render them as a badge in the transcript viewer
• CSS: .evt-item.type-thinking at line 2072 — add collapsed state styles