fix: Prevent SSH idle disconnects via proper keepalive wiring

[inureyes]

Summary

Idle SSH sessions disconnected inconsistently — sometimes after a few minutes, sometimes after ~10 minutes. Fixes three underlying issues:

1. russh inactivity_timeout ceiling — to_russh_config() inherited russh's 10-minute default via ..Default::default(), imposing a hard upper bound regardless of SSH keepalive. Now set to None when keepalive is enabled.
2. No TCP SO_KEEPALIVE — the underlying TCP socket had no kernel-level keepalive, so NAT/firewall conntrack entries could silently expire between SSH keepalive packets. connect_with_config now builds the TcpStream manually, applies socket2::TcpKeepalive derived from the SSH keepalive config, and hands it to russh::client::connect_stream.
3. Exec-mode config dropped — SshConnectionConfig was threaded through the executor but the ConnectionConfig field was #[allow(dead_code)], so user-configured server_alive_interval never reached Client::connect_with_ssh_config. The field is now wired end-to-end through connect_direct / connect_via_jump_hosts / the jump chain.

Changes

• src/ssh/tokio_client/connection.rs: override inactivity_timeout, add to_tcp_keepalive(), rewrite connect_with_config around connect_stream with SO_KEEPALIVE.
• src/ssh/client/{config,connection,command,file_transfer}.rs: thread ssh_connection_config through the exec path.
• src/executor/{connection_manager,parallel}.rs: forward SshConnectionConfig into ConnectionConfig; remove dead_code.
• Cargo.toml: add socket2 = "0.6" (already transitive via tokio).

Test plan

• cargo build / cargo clippy --all-targets clean
• cargo test --lib — 1193 passing (3 pre-existing keychain tests fail due to macOS keychain auth prompt, unrelated)
• Manual: leave interactive session idle >10 min on a NAT'd path
• Manual: verify server_alive_interval in ~/.config/bssh/config.yaml now affects exec mode