Session picker should show continuation/chain linkage between transcripts

Feature request

When a session runs out of context and continues into a new transcript, the two files become separate entries in claude --resume (and the claude.ai/code list) with no visible relationship. Long-running work produces chains of 3–5 transcript files that look like unrelated sessions.

What the transcripts show today

• A continuation re-stamps the carried-over history with its own sessionId — there is no parentSessionId/origin field linking a chunk to its predecessor.
• The only de-facto linkage signals are accidental: identical first prompt/birth timestamp, and the custom-title record — which propagates forward only. Rename a session before it continues and the whole chain shares the name; rename after and earlier chunks keep stale auto-titles, so the chain looks unrelated in the picker.
• /rewind cannot cross the boundary (by design), so finding the earlier chunk is exactly the moment you need the linkage — and today that means grepping ~/.claude/projects/ for matching first timestamps.

Ask

1. Record explicit chain metadata at continuation time (e.g. parentSessionId in the new transcript's header records).
2. Use it in the resume picker and the claude.ai/code session list: group a chain under one entry (expandable), or at least badge continuations ("continued from …") and sort them together.
3. Propagate a rename across the whole chain (or offer to).

Why it matters

Real example that prompted this: a two-week feature produced a 4-file chain plus a sibling worktree chain. After the worktree was torn down, reconstructing which transcript held which era of the work required inspecting file headers by hand; the picker showed five unrelated-looking sessions.