app-server: clean up delegated serve child when dispatcher exits

[Hmbown]

Follow-up from v0.8.61 app-server release verification

codewhale app-server --http / --mobile delegates to the sibling codewhale-tui serve path. The dispatcher waits with Command::status(), but killing the dispatcher can leave the delegated codewhale-tui listener alive and reparented.

This was treated as a v0.8.62 teardown hardening follow-up, not a v0.8.61 release blocker, because normal foreground shutdown and explicit listener cleanup worked during verification.

Expected direction:

• Add a controlled shutdown/cleanup path for delegated long-running server children.
• Add a regression smoke that starts a delegated runtime API listener on a loopback free port, terminates the dispatcher, and asserts no listener remains on that port.

[Hmbown]

Retargeting into v0.8.63 from the release triage pass. This belongs in the same lane as the config/sub-agent/runtime-control work because it affects user trust, release stability, or first-run/runtime behavior.

[Hmbown]

Partial v0.8.63 update from PR #3347:

Commit 11df4e5 landed the catchable-signal teardown path for delegated serve / app-server children. It supervises the delegated child, forwards Ctrl+C / SIGTERM / SIGHUP where supported, kills and reaps the child before the dispatcher exits, and keeps the signal-style exit code.

That covers the normal controlled-shutdown path, but the commit intentionally leaves the uncatchable dispatcher-death case open: SIGKILL still cannot run cleanup code, and needs OS-level follow-up such as Linux PR_SET_PDEATHSIG or Windows Job Objects.

Because that remaining edge is real and explicitly out of scope for the landed slice, I am keeping this issue open and retargeting the remaining work to v0.8.64 rather than closing it from the partial fix.