fix: abort process reader/waiter tasks on cleanup to prevent hang

[jbj338033]

Summary

ProcessRegistry spawned reader/waiter tasks without keeping their JoinHandles, so if a group kill raced or missed, a leaked child kept those tasks (and its inherited stdout/stderr pipes) alive — stalling process shutdown. This was the root cause of the intermittent 48-minute CI hang on test (ubuntu-latest) and the Windows stdin_write_reaches_process failure (findstr never echoes until stdin EOF).

Task handles are now stored on Entry and aborted on drop; kill also closes stdin to send EOF; shutdown_all drains entries to guarantee cleanup. Adds shutdown_all_terminates_running_processes as a regression test.

Verification

• cargo test -p goat-agent — passed (110 tests)
• cargo clippy --workspace --all-targets -- -D warnings — passed
• cargo fmt --all --check — passed
• no orphan sleep 30 processes after kill/shutdown tests — verified