fix(runtime): preserve client capability tool outcomes

[testikun]

Summary

Computer Use refusals and explicit stops could return model-visible text without a call-failure fact. Desktop's Client Capability projection then dropped even existing result status, so Host settlement, T2, telemetry, and activity views could record a failure as success. This PR carries an explicit success | error | aborted outcome from CU/native projection through the Client Capability wire and Host tool settlement, then through durable events, continuity, UI, CLI, and ACP replay. Ordinary MCP and business JSON results retain their existing semantics; failed process observations remain successful tool calls.

Client Capability results now require the outcome and the Host compatibility epoch moves from 161 to 162. Old stored records without the field remain readable; this does not promise that an older binary can read newly written records. No old records are rewritten. The change stays within the bounded outcome contract, without a general ToolResult/plugin migration.

Fixes #5449
Refs #4909

Verification

• npm run build, npm run typecheck, npm run lint, npm run format:check, npm run check:stale, and npm run check:asf-headers passed.
• Runtime Host full dist suite on Node 24 with --test-concurrency=1: 1,952 passed, 12 skipped, 0 failed. Targeted CU, direct/chunked UDS, T2/SQLite reopen, continuity, UI, CLI/ACP, and Desktop native projection regressions passed. The other workspaces passed in the serial workspace run.
• The default workspace runner twice failed only the unchanged session-catalog-two-client-uds CAS race (committed, committed versus committed, revision_conflict); that file passes alone and the complete Runtime Host suite passes with one test file at a time. No CAS code or assertions were changed.
• No real OS permissions, external network provider, or model execution were exercised. Existing controlled-transport evidence is not presented as a new real-machine screenshot.

AI use

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

Tool(s) and scope: OpenAI Codex implemented the bounded result contract, cross-layer propagation, regression tests, and verification. The commit carries a Generated-by trailer.

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