Fix SSE reliability and message delivery for production environments

[chriswritescode-dev]

Summary

This PR addresses SSE connection reliability issues, particularly for deployments behind Cloudflare or similar proxies that terminate idle connections. It also switches to synchronous message delivery to prevent message desync during complex operations.

Changes

Backend:

• Add SSE heartbeat mechanism (60-second interval) to keep connections alive and prevent Cloudflare 524 timeout errors

Frontend:

• Switch from /prompt_async to synchronous /message endpoint for reliable message delivery
• Improve optimistic update handling with proper query cancellation before updates
• Handle network errors gracefully (don't clear optimistic messages on network failures)
• Fix SSE event handling to invalidate queries instead of setting empty data when cache is missing
• Add heartbeat event listener to acknowledge keep-alive signals

Shared:

• Add HEARTBEAT_INTERVAL_MS constant (60000ms) to SSE defaults

Why These Changes

1. Cloudflare 524 Timeout: Without regular activity, Cloudflare terminates connections after ~100 seconds of inactivity. The heartbeat prevents this.

2. Message Desync: Using the async endpoint could cause the UI to miss authoritative response data, especially during sub-agent tasks. The sync endpoint matches OpenCode TUI behavior and returns the confirmed message.

3. Race Conditions: Proper query cancellation before optimistic updates prevents React Query race conditions.