In PR #1029 we removed the histograms from the command_processor_process metrics because of the extremely high cardinality that was causing problems for Grafana dashboards and metrics forwarding.

The high cardinality was caused by the multiplicity of tags we supported including command, tenant, sort type, error, error class, error_code, and vector_enabled. When adding histogram buckets to this, it potentially results in a huge number of series.

However it would still be useful to be able to track latency by command for debugging purposes. We should add a new metric called command_processor_latency that is a histogram metric tagged by command and tenant only.