feat(recap): add durable native Session recap plugin

[me2seeks]

Summary

Adds statically linked maka.session-recap with a manual Desktop Session Inspector contribution and a Session-bound Remote endpoint.

• Read authorized, fixed-watermark conversation history and generate a one-sentence recap through the Session's selected Host model.
• Atomically store operation intent and latest pointer before dispatch. Reusing an operation ID reads its receipt, including after a lost reply or restart; uncertain requests are never automatically redispatched.
• Keep recap results in plugin storage, recheck source history access on cached reads, and drain cancelled model work. Older completions cannot replace newer requests.

This provides manual generation and durable inspection. Automatic idle triggers, Daily review and the legacy TS recap RPC are outside this plugin. Public history currently offers text, not TS structured tool-outcome projections; oversized scans explicitly refuse before model dispatch.

Verification

• cargo test -p maka-session-recap --lib: 5 passed, covering durable retry/restart, cached access refusal, concurrency, unknown results, cancellation, privacy and history bounds.
• cargo test -p maka-runtime-host --test integration session_recap_plugin: 1 passed through actual Session/model/Remote binding, duplicate prevention, retirement and Host restart.
• Focused Clippy with warnings denied, Rust formatting, dependency license/inventory check, Client strict TypeScript check, Biome lint/format and diff whitespace check passed.
• Browser exercised the actual Client in an isolated mock-Remote preview. After a simulated lost reply, “Check original request” reused exactly the same operation ID and displayed the recovered recap. This was not a full Electron test.
• GPT-6 Luna independently audited lifecycle/idempotency and verified that Session retirement tombstones prevent ID reuse. Explicit scan-limit refusal is covered by tests.

No paid external model or repository-wide suite was run.

AI use

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Codex implementation and verification; GPT-6 Luna design and boundary audit. Commits include Generated-by: Codex.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No