fix(del): pass --force to codex delete so it runs headless

[jarvan1]

Problem

Deleting a codex session from the picker removed the rollout .jsonl file but left codex's index row behind (the threads row in ~/.codex/state_*.sqlite), so every deleted codex session leaked a dead entry that still shows up in codex resume.

Root cause

The picker runs codex delete <id> via exec.Command from a non-interactive context (no TTY). Recent codex-cli refuses there:

Error: cannot confirm session deletion without an interactive terminal; rerun with --force and a session UUID

Our native step is best-effort (_ = cmd.Run()), so this failure was swallowed — the file got removed by the fallback, masking that the index prune never ran.

Fix

Pass --force so the native prune actually runs headless. File removal still backs it up if the command is missing.

- p.nativeCmd = []string{"codex", "delete", s.ID}
+ p.nativeCmd = []string{"codex", "delete", "--force", s.ID}

Verification

• Reproduced on codex-cli 0.142.5: plain codex delete <uuid> → refuses without TTY; codex delete --force <uuid> → Deleted session <uuid>. and the threads row is gone.
• Reconciled the local index vs rollout files afterward: 16 index rows == 16 files, zero orphans.
• go build ./... && go vet ./... && go test ./... all pass; del test updated to expect the --force arg.

🤖 Generated with Claude Code