fix(server): drain Handle::data backlog before select!

[Yaminyam]

Summary

tokio::select! in the server session loop doesn't always wake up promptly for messages produced by spawned tasks via Handle::data() when the internal mpsc channel is being filled faster than the select cycle drains it. In interactive PTY servers where a separate task pumps shell output through Handle::data(), this shows up as the client missing chunks of output until some other event happens to wake select.

This PR drains the receiver with try_recv() before entering select!, with a 64-message cap so select still gets to run between batches and client input (Ctrl+C, window resize, …) stays responsive.

Reproduction

Any russh-based SSH server with a PTY session whose shell output is generated from a spawned task pushing via Handle::data() faster than the session loop's select cycle reads from its mpsc channel — e.g. yes | head -n 100000 over an interactive PTY. The client sees periodic stalls / dropped output until further activity.

Change

In run_inner (server session loop), before the existing tokio::select!, batch-drain up to MAX_MESSAGES_PER_BATCH = 64 messages from self.receiver with try_recv() and dispatch through the same handlers select! would have used: data / extended_data / eof / close / channel_success / channel_failure / xon_xoff_request / exit_status_request / exit_signal_request / WindowAdjusted / ChannelOpen{Agent,Session,DirectTcpip,...}. Gated on !self.kex.active() so the rekey state machinery isn't perturbed.

The cap exists so high-throughput output can't starve client input — once a batch is full we fall through to select!, which picks up any client-side event (input, channel close, keepalive timeout) before the next batch.

Origin

This is the same fix bssh has been carrying as crates/bssh-russh/patches/handle-data-fix.patch. Its README pins at inureyes/russh:fix/handle-data-from-spawned-tasks and notes "When merged upstream, this fork will be deprecated" — this PR is the upstream half of that contract. Original author of the fix: @inureyes (bssh maintainer).

Test plan

• cargo build -p russh → clean
• Manual PTY soak test recommended before merge: spawn-task-driven Handle::data() server, confirm no output stalls under yes | head -n 100000-style load.

[Eugeny]

Thanks for bringing it to my attention - I've added my own cleaner fix in #731

@all-contributors please add @inureyes for code

[allcontributors]

@Eugeny

I've put up a pull request to add @inureyes! 🎉