Per-turn token usage, context occupancy, and cost display

[harrryyd]

Problem

After each turn completes, the user has no visible breakdown of:

• Tokens used in that turn
• Context window occupancy (% of max)
• Approximate cost of the turn

Proposal

Add a per-turn summary UI that appears after each turn completes, showing:

• Tokens used: input + output for the turn
• Context window: percentage bar (usedTokens / maxTokens)
• Cost estimate: based on model pricing × token counts

This data is already tracked internally:

• ClaudeAdapter now captures per-request usage from message_start/message_delta stream events
• thread.token-usage.updated events carry usedTokens, lastUsedTokens, totalProcessedTokens, maxTokens
• Codex and OpenCode adapters emit similar usage events

Considerations

• Should this be inline in the message timeline (after each assistant response) or a summary panel?
• Should cost be shown by default or opt-in?
• Model pricing table needed (may already exist in provider config?)

Related

• Spotted during review of feat: provider compaction + context trim + archive-and-new + review flow #108
• ClaudeAdapter per-request usage tracking: commits 8955033, 7b9cef7