Handle SIGTERM gracefully in server and master

[KaiqiJinWow]

Why

Kubernetes stops a pod by sending SIGTERM. The server and master previously waited only for Ctrl+C, so a rollout or scale-down could terminate either process before Axum drained active connections. For the server, this also meant state.shutdown() was never reached to close resident writers cleanly.

flowchart LR
    subgraph before ["Before"]
        direction TB
        ctrl_before["Ctrl+C / SIGINT"] --> graceful_before["Axum graceful shutdown"]
        term_before["SIGTERM"] --> exit_before["Process exits immediately"]
    end

    subgraph after ["After"]
        direction TB
        ctrl_after["Ctrl+C / SIGINT"] --> select_after["tokio::select!"]
        term_after["SIGTERM"] --> select_after
        select_after --> graceful_after["Axum graceful shutdown"]
        graceful_after --> drain_after["Drain active connections"]
        drain_after --> close_after["state.shutdown(): close writers"]
    end

Loading

For example, during a Kubernetes rolling deployment:

1. Kubernetes sends SIGTERM to the old pod.
2. The process now stops accepting new requests and lets in-flight requests finish.
3. The server closes its writers before exiting with status 0.

What changed

• Wait for either SIGINT/Ctrl+C or SIGTERM on Unix in both binaries.
• Keep the existing Ctrl+C behavior on non-Unix platforms.
• Log which signal initiated shutdown.
• Verify that the Python server fixture exits successfully after Popen.terminate() sends SIGTERM on POSIX.

Testing

• Server test suite: 80 passed.
• Clippy for server and master, including all targets.
• Rust formatting and Python lint checks.
• Real-server SIGTERM smoke test: exited with status 0.