bridge: restore fork lineage on cold thread lists

[0xSero]

Revival of the closed #218 (branch deleted). Fixes the cold-thread/list half of #150: the fork rollout survived process restart and thread/read returned its parent, but cold thread/list discarded forkedFromId in the local thread-store list projection. The patch restores lineage from the rollout session metadata.

Verification:

• git apply --check of patches/codex/thread-list-fork-lineage.patch against a clean export of the pinned codex (13595c36) — applies cleanly; the submodule pin has not moved since the original draft.
• Sync-codex wiring (Makefile resolved to main's wildcard patch discovery; the patch is picked up automatically).
• Full physical-device acceptance from Forked chat/thread becomes inaccessible after leaving session #150 (fork → follow up → leave → force restart → reopen from Sessions/Forks → continue on both iPhone This-Device and remote-VM paths) remains the issue's acceptance gate.

[0xSero]

Follow-ups from the review, pushed as 411e6a26..1e4239c9:

1. CI now runs the new test. The shared job applies the patch set (sync-codex.sh --recorded-gitlink) and then runs cargo test -p codex-thread-store, so cold_list_threads_preserves_fork_lineage_from_rollout executes on every PR instead of only locally.

2. Per-thread re-reads removed. Lineage is now carried through the rollout list projection instead of reopening each rollout file in list_threads: ThreadItem gains forked_from_id (filled from session meta during the head read), stored_thread_from_rollout_item forwards it, and fill_missing_thread_item_metadata merges state-DB metadata values in. thread-list-fork-lineage.patch was regenerated accordingly. The whole patch sequence was verified by resetting the submodule pristine and re-applying every patch from the committed files — the re-applied tree is byte-identical to the tested state.

3. Self-parents / cycles / missing parents. Left as-is intentionally: upstream fork always mints a fresh thread id, and with the projection path the parent id comes straight from the fork's own session meta. A missing parent (deleted, archived, or on another page) yields a fork entry whose parent id may not be present in the same page — same exposure as the previous per-read approach; deeper UX handling is tracked with Forked chat/thread becomes inaccessible after leaving session #150.

4. Silent error nit — evaporated with the re-read block: list_threads no longer does per-thread file I/O for lineage.

5. Test gaps. The cold-list test now also asserts a non-fork thread keeps forked_from_id: None. The state-db-only and id-mismatch gaps are moot — no per-thread file reads remain, and the projection is per-file by construction. A mobile-side reducer test asserting a cold list result sets is_fork in the session summary is still open; noted on Forked chat/thread becomes inaccessible after leaving session #150.

Verified locally: cargo test -p codex-thread-store 73/73 pass, cargo check -p codex-rollout --all-targets clean.