Log dropped subscription events

[peter941221]

1. Failure mechanism

Subscription delivery already uses non-blocking sends and the public docs already warn that slow consumers can lose events. Right now, when a subscriber channel fills up, subscription_manager drops the event silently. That leaves you with no signal that a subscription fell behind or which kind of event was lost.

2. Semantic change

This adds a warn log when a job or queue subscription event is dropped because the subscriber buffer is full.

The log includes event_kind so you can tell what was lost.

3. Why this design

I kept the existing delivery semantics. This does not add retries, change buffering, or block producers behind a slow subscriber. It only turns a known lossy path into something observable.

4. Testing

5. Added Test_SubscriptionManager/LogsDroppedQueueEvents in subscription_manager_test.go.

6. Ran go test . -run Test_SubscriptionManager/LogsDroppedQueueEvents -count=1.

7. Ran go test . -run TestDoesNotExist -count=1 as a compile/load smoke check.

[brandur]

Yeah, this isn't a bad idea. Want to add a changelog?

[peter941221]

Added a changelog entry under Unreleased -> Changed.