Metrics for merge buffer usage

Would be nice to have two metrics for merge buffers, so operators can know if they are frequently maxed out on merge buffers.

- number of merge buffers currently allocated. I imagine it'd emit the instantaneous number of merge buffers used, once a minute. So it might miss brief spikes but should catch prolonged ones.
- amount of time spent waiting for a merge buffer. I imagine this would be per query.

The issue names no files or tests; start by locating the Java paths responsible for merge-buffer allocation and waiting. Done means operators can observe current merge-buffer usage and per-query time spent waiting, with behavior matching the requested sampling and aggregation.

Written by the indexing model from the issue text.