fix(stream-bridge): guarantee END sentinel delivery to prevent resource leaks

[JasonOA888]

Fixes #1689

Problem

When the stream bridge queue is full (256 events), discards the END sentinel after a 30s timeout. Since only exits upon receiving , losing it causes:

• SSE connections to hang indefinitely
• and never cleaned up
• Accumulating resource leak under sustained load (long agent runs with slow consumers)

Root Cause

used the same strategy as . But END sentinel is not a normal event — it is the only way subscribe() knows to exit.

Fix

• : use instead of — drops events immediately when queue is full, no 30s stall per dropped event
• : retry loop that blocks until queue has space, with periodic warning logs every 10 seconds. END sentinel is never dropped.

Testing

• Syntax validated via ast.parse
• Single file change, no new dependencies

[JasonOA888]

Closing this PR — the code it modifies no longer exists.

The original StreamBridge used asyncio.Queue with a put(timeout=30s) strategy that could drop the END sentinel. This was fixed by #1695 (merged), and then the entire StreamBridge was rewritten in #1403 to use a list + asyncio.Condition architecture (MemoryStreamBridge) that does not have this issue.

The current publish_end sets stream.ended = True and notifies waiters — there is no queue to fill, no timeout to expire, and no sentinel to drop.