[kunalgoesbyken]

Closes #1319

Problem

railway connect <db> --tunnel-only only closed its tunnel on Ctrl+C. Stopping it any other way (kill, pkill, systemd, closing the terminal) left the ssh -N -L child running under PID 1, still forwarding the local port to the database.

Mechanism

• spawn_native_ssh_forward (src/commands/ssh/native.rs) puts ssh in its own process group (process_group(0)), so signals sent to railway never reach it. Only ForwardGuard::drop kills it.
• run_tunnel_only (src/commands/connect.rs) awaited only tokio::signal::ctrl_c(). SIGTERM and SIGHUP took the default action and ended the process without unwinding, so the guard was never dropped.

Fix

• run_tunnel_only now awaits wait_for_shutdown_signal(). On Unix that selects over Ctrl+C, SIGTERM and SIGHUP; elsewhere it stays Ctrl+C. Any of them returns normally, the guard drops and ssh is killed, the same as the Ctrl+C path today.
• run_ssh_connect is unchanged. It deliberately ignores SIGINT so the database client can use it.
• No PR_SET_PDEATHSIG: it is Linux-only and fires when the spawning thread exits, not the process, which is fragile under tokio. SIGKILL is out of scope.

Verification

• test_shutdown_signal_resolves_on_terminate_and_hangup (Unix): registers the handlers, checks the future is still pending, raises SIGTERM and then SIGHUP, and checks each one resolves it.
• cargo fmt --all -- --check and cargo clippy --all-targets --all-features, on Windows and for x86_64-unknown-linux-gnu.
• cargo test: all commands::connect tests pass.

This needs a release/patch label, which I can't set from a fork.

[kunalgoesbyken]

This is a bug fix to railway connect --tunnel-only, so release/patch should fit for the label check. I can't add labels from a fork.