mobile: persist thread plan mode across relaunch

[0xSero]

Purpose

Fixes #100 — plan mode state is lost after the app crash/relaunch. Revival of the closed #220 approach, rebuilt on current main. Reported and analyzed by @julianpistorius — thank you for the detailed reproduction and the persistence-direction suggestion.

Key changes

• Persist non-default per-thread collaboration mode in a bounded, Rust-owned thread_modes.json (shared Rust, one store file — no platform-side parsing).
• Restore the persisted mode through the canonical reducer on list, lifecycle, and hydrated snapshots.
• Reconstruct the "Implement Plan" affordance on cold hydration only when the latest proposed plan has no later user turn (restore_plan_implementation_prompt_from_history, wired into upsert_thread_snapshot with a properly scoped existing-thread lookup).
• Explicit Default keeps clearing the stored mode; dismissed prompts are not resurrected.

Verification

• cargo test -p codex-mobile-client -p codex-slingshot on the rebased branch (current main 0e843080): 806 lib tests + all slingshot suites, 0 failures — includes three new plan-persistence tests (restore-on-cold-hydration, dismissed-prompt stays dismissed, user-turn-after-plan suppresses restore) plus the item-completed / turn fixture repairs.
• upstream::ItemCompletedNotification and upstream::Turn test fixtures updated for the upstream struct fields (completed_at_ms, items_view).

Notes

• Rebase dropped an unrelated stale docs commit from the old branch (it described an alleycat pin form that no longer matches main).
• Follow-ups for Forked chat/thread becomes inaccessible after leaving session #150's fork-lineage UI work are tracked separately.