Implement tracing and client metrics

[Ladicek]

Motivation:

• make sure that the Redis client calls back on the original context
  • this used to be the case for the clustered clients, but nor for the standalone client, which is fixed here
• make connection pooling explicit in the code
  • the internal structure of connection handling is more obvious by using the PooledRedisConnection type explicitly
  • this even revealed a bug in RedisSentinelClient which would lead to class cast errors in some failure scenarios
• implement tracing and client metrics
  • this is done similarly to how the Vert.x SQL client does it, which is a little less precise, but easier to understand

Resolves #357

[Ladicek]

If you agree, I'll submit a backport PR against the 4.x branch after this is merged.

[Ladicek]

No idea why the action failed -- the Run tests step succeeds, but then Publish Unit Test Result fails with Error: Resource not accessible by integration :shrug:

[Ladicek]

Rebased.

[vietj]

@Ladicek can you rebase your PR so we can review it next ?

[Ladicek]

Rebased.

[Ladicek]

Added one more commit that improves the replication test, although I'm not sure if that isn't really a deficiency of the replication client :shrug:

[Ladicek]

Removed useless imports, thanks for noticing!

[vietj]

@Ladicek you can merge and backport the PR

[Ladicek]

Backported to 4.x in #398.