SIGTERM is unhandled, so every managed restart is a hard kill

The server does not handle SIGTERM, so every managed restart is a hard kill with connections in flight.

`serve`'s own doc comment in `rs/cyphr-server/src/lib.rs` says it "blocks until SIGTERM/SIGINT". `shutdown_signal()` awaits `tokio::signal::ctrl_c()` only, which is SIGINT. SIGTERM — what systemd, Docker, Kubernetes and a bare `kill` all send by default — is unhandled, so the process dies on the default disposition without draining.

## Reproduction

Send SIGINT:

```
shutdown signal received, draining connections
server stopped
```

Send SIGTERM: neither line appears. The last entry in the log is `server started`.

## Impact

The store survives — restarting after a SIGTERM kill opens cleanly with no stale lock — so the damage is bounded to requests in flight. But that is every `systemctl restart`, every container stop, and every rolling deployment.

The doc comment claiming both signals makes it worse: an operator reading the source concludes draining is handled.

Start in rs/cyphr-server/src/lib.rs, reading the serve doc comment and the shutdown_signal() implementation. Reproduce the difference between SIGINT and SIGTERM, then verify that SIGTERM produces the draining and stopped log messages and no longer hard-kills the server with connections in flight.

Written by the indexing model from the issue text.