In #21280, we're adding forwarding of boundary logs from boundary to coderd logs. During implementation and code review it's apparent that we may drop batches of logs in boundary before they are sent to the agent, including:

1. Internal log buffers may become full waiting for batch I/O
2. AI agent network request bursts could exceed the buffer size

This ticket captures adding visibility for if and when these logs are dropped.

Consider having boundary tell the agent when this happens, and how many batches were dropped. The agent could then export a metric for observability. Consider exporting metrics in such a way that we will know why batches were dropped.

coder/boundary#124 (comment)