fix(orchestration): fail direct dispatches on agent exit

[tianrking]

Summary

• mark a task as failed when a directly dispatched worker exits and no coordinator run exists
• preserve the exit reason on the task instead of leaving it indistinguishable from an untouched ready task
• add a runtime regression test for the direct-dispatch path

Root cause

failDispatch correctly returns a task to ready for coordinator-managed retries. A direct orca orchestration dispatch has no coordinator loop to consume that retry, so an agent exit could leave work silently stranded in ready forever.

Impact

Coordinator-managed dispatches retain their existing retry and escalation behavior. Directly dispatched tasks now become visible failures that can be deliberately retried by the operator.

Validation

• node_modules/.bin/vitest.cmd run --config config/vitest.config.ts src/main/runtime/orca-runtime.test.ts --testNamePattern 'marks a direct dispatch failed'
• node_modules/.bin/oxfmt.cmd --check src/main/runtime/orca-runtime.ts src/main/runtime/orca-runtime.test.ts
• node_modules/.bin/oxlint.cmd src/main/runtime/orca-runtime.ts src/main/runtime/orca-runtime.test.ts

The full runtime test file passed 710 tests; 4 unrelated tests assume a POSIX /tmp path and fail on this Windows host before exercising this change.

Fixes #8984