Reduce metrics cardinality replication.TaskStore

[vytautas-karpavicius]

What changed?
Emit replication cache size as timer, to get a percentile view instead of metric per each shard.

Why?
Turns out, having metric per shard is very expensive. Dashboard become too sluggish to work with.
Keep higher level metrics only and log warning when cache is full, for debugging specific shards.

How did you test it?

Potential risks

Release notes

Documentation Changes

[coveralls]

Pull Request Test Coverage Report for Build 0182f365-4f4e-4bbb-8547-57534eeb5660

• 10 of 10 (100.0%) changed or added relevant lines in 1 file are covered.
• 127 unchanged lines in 13 files lost coverage.
• Overall coverage decreased (-0.02%) to 57.78%

---

Files with Coverage Reduction	New Missed Lines	%
common/task/weightedRoundRobinTaskScheduler.go	1	88.6%
common/backoff/retry.go	2	92.11%
common/persistence/statsComputer.go	3	94.64%
common/task/fifoTaskScheduler.go	3	84.54%
common/persistence/serialization/parser.go	4	65.41%
common/persistence/serialization/thrift_decoder.go	4	61.22%
service/history/task/transfer_standby_task_executor.go	4	86.79%
common/persistence/nosql/nosqlplugin/cassandra/workflow.go	6	56.99%
common/persistence/nosql/nosqlplugin/cassandra/workflowUtils.go	11	74.22%
common/persistence/sql/workflowStateMaps.go	11	85.56%

Totals
Change from base Build 0182eb95-a47b-4b48-8f18-de9548f005fc:	-0.02%
Covered Lines:	84371
Relevant Lines:	146020

---

💛 - Coveralls