bug fix: use buffered message sender from v2 in v1

[ryantm]

Why

• We have a problem with the connection between agent and container sometimes hanging.
• When I run our agent tests with debugging levels to the max it does not seem to trigger over many runs.
• I passed this hypothesis of messages not flushing to Claude and it suggested using this buffered message sender we already have.
• Note, since I haven't reproduced the original hanging problem locally, I cannot confirm that this fix fixes that. Right now our best way to confirm that is to ship this change to CI and see if the hangs stop.

What changed

• Use the buffered message sender from river v2 in v1

Test plan

• Tests still pass locally.
• I tried connecting this version of the package to my local agent setup and it works fine

[macroscopeapp]

Route all replit_river.session.Session outbound messages through the v2 buffered_message_sender background task to replace immediate sends in v1

This change starts a background buffered sender in Session and removes immediate send paths, routing all enqueued messages through the v2 sender.

• Start Session._start_buffered_message_sender to schedule buffered_message_sender with session-specific callbacks in session.py
• Remove direct send logic from Session.send and Session.replace_with_new_websocket, relying on the background task in session.py
• Drop imports and exception handling tied to immediate send paths in session.py

📍Where to Start

Start with the Session._start_buffered_message_sender helper in session.py, then review its callbacks and how Session.send now only enqueues into the buffer.

---

Macroscope summarized 701e02b.

[ryantm]

This did not fix the problem after testing in CI.