feat(hostlink): split TaskJob into shared runner and executor, route polling through enqueue, separate durable start from transport start

[iAziz786]

Summary

Refactor TaskJob to decouple task receipt from execution so polling fallback and WebSocket delivery share one runner queue. Also splits the started lifecycle so durable local-state persistence happens before process launch, separate from task.started transport emission.

Changes

• Shared runner queue: polling trigger no longer calls processTask directly; every task passes through Enqueue into one sequential runner loop
• Executor extraction: shell launch, output capture, and result reporting live in taskExecutor, not TaskJob; polling and WS tasks share the same execution path
• In-memory duplicate suppression: reserveAttempt/finishAttempt keyed by task_id + execution_attempt_id
• Started lifecycle split: ResultChannel.RecordStarted persists running state before process launch; SendStarted only emits the transport message
• WebSocket client: RecordStarted added, SendStarted slimmed to transport-only
• Integration test timing: all TaskJobReporter integration tests updated for async runner model

Testing

go test ./app/jobs/taskjob ./app/services/wsclient ./app/services/localtaskstore ./internal/wsprotocol
go test -tags integration ./test/integration -run 'TaskJobReporter'

Note

This PR includes the uncommitted ws-task-delivery changes that were in the working tree. The effective runner-refactor change is best reviewed as the diff from the ws-push-task-delivery base.