Preserve process facts when the remote Holder owner disappears

[cristicretu]

Remote inspect used to rewrite a missing Holder lock into Exited { code: None, signal: None }, so reconnect and scripts could accept an exit that was never observed. It now returns exit status 1 with {"error":"holder_unavailable"}, preserves the stored process state, and keeps recorded exit facts intact.

The Engine retains the typed error as NotConnected. Existing successful responses are unchanged; older Engines already reject nonzero Helper RPCs. No attachment, controller change, or process signal is added. kill semantics are a separate follow-up.

Validation

• Real Helper subprocess with a disposable live child: typed owner failure, unchanged state bytes, child remains alive, authenticated facts only, recorded exit126 preserved.
• Real Holder crash still cleans up the Agent; inspection reports owner failure without inventing exit facts.
• State/lock/incarnation fixtures and Engine old/unknown/truncated error decoding.
• Regression fails with the old success/fabricated-exit behavior.
• Full workspace: 1,686 passed / 36 ignored, formatting, strict workspace Clippy and release build passed.

CLI/test evidence

Rendered synthetic fixtures and test results; no user session data or GUI claim.

Stacked on #340 for the shared process identity work.

[cloudflare-workers-and-pages]

Deploying diri with    Cloudflare Pages

Latest commit:	3cfaf08
Status:	✅  Deploy successful!
Preview URL:	https://59f5a403.diri-53j.pages.dev
Branch Preview URL:	https://fix-remote-inspection-owner.diri-53j.pages.dev

View logs