fix(core): register signal handler before startup work and handshake

[bug-ops]

Summary

• run_stdio registered its SIGTERM/SIGINT handler only after mcp_server.serve(..)'s MCP initialize handshake resolved, and serve_with ran config validation, workspace-root heuristics, and spawn_lsp_servers_background before run_stdio ever got a chance to register one. A signal delivered anywhere in that window fell through to the OS's default disposition (immediate kill), skipping Translator::shutdown_servers and risking an orphaned LSP child process mid-spawn.
• ShutdownSignal is now constructed once, as the first statement in serve_with, before any startup work runs, and moved by value into whichever transport runs; run_stdio races it against the handshake itself, then reuses the same instance in its existing post-handshake select!.
• Also fixes a related gap found during review: SIGINT was previously re-registered via a fresh one-shot tokio::signal::ctrl_c() on every wait, which could silently lose a signal delivered while a different select! branch was being polled. ShutdownSignal now holds a persistent listener per signal kind for its entire lifetime.

Fixes #318.

Test plan

• cargo +nightly fmt --all -- --check
• cargo clippy --all-targets --all-features --workspace -- -D warnings
• cargo nextest run --workspace --all-features --lib --bins (643 passed)
• RUSTFLAGS="-D warnings" RUSTDOCFLAGS="--deny rustdoc::broken_intra_doc_links" cargo doc --no-deps --workspace --all-features
• cargo test --doc --workspace --all-features
• New regression test test_e2e_sigterm_exits_promptly_during_handshake_wait, targeting the actual pre-fix window (SIGTERM sent before initialize() is ever called): 5/5 failures against pre-fix code, 10/10 passes against the fix
• Existing test_e2e_sigterm_exits_promptly_while_client_stdin_open (mcpls process does not exit on SIGTERM/SIGINT while a stdio client is still connected #308 regression) still passes with its artificial delay removed
• Empirically verified 0 orphaned LSP child processes across 8 kill-delay timings (0.001s-0.300s) against a stand-in long-running LSP server