conversation resume: extend native-id capture coverage + robustness

Deferred from `docs/design/2026-06-25-agent-conversation-resume-design.md`. Today `forcesID` covers claude only and `scrapesID` covers codex only; other agents fall back to generic/fresh resume.

- [ ] **OpenCode force** — after verifying `--session` accepts a brand-new UUID-format id, widen `forcesID` to `opencode` and set `opencode.args = ["--session", "{agent_session_id}"]`.
- [ ] **Agy locator** — reverse-engineer Agy's transcript layout (likely `~/.gemini`) and add a locator so its native id can be scraped/pinned.
- [ ] **Cursor capture** — capture Cursor's native id via a `~/.cursor/projects/` locator or the `sessionStart` hook, removing reliance on generic cwd-scoped `resume`.
- [ ] **Verify-before-resume** — probe that the transcript still exists before id-pinned resume; fall back to fresh start with a warning if it's gone.
- [ ] **Capture provenance in `gr list`** — record how the id was obtained (forced/scraped/migrated) and surface it so users know whether resume is reliable.

Read docs/design/2026-06-25-agent-conversation-resume-design.md, then locate the forcesID and scrapesID handling and the gr list path. Verify OpenCode session behavior, investigate the stated Agy and Cursor transcript or hook locations, and trace resume when a transcript is missing. Done means the listed agents have native-id capture or forcing, missing transcripts fall back with a warning, and gr list shows id provenance.

Written by the indexing model from the issue text.