Partition-scoped metrics for the SQL Server connector

The connector task working with multiple partitions should expose metrics in a different format. The existing metrics logically can be divided into two categories:

1. Task-scoped metrics. For instance, QueueTotalCapacity. While processing multiple source databases, the task will use a single queue for the messages sourced from all partitions.
2. Partition-scoped metrics. For instance, MilliSecondsBehindSource. Such metrics should be reported individually for each source database which is currently impossible.

The introduction of partition-scoped metrics is necessary for the following reasons:

1. Being able to operate the connector. Without a partition identifier, the partition-scoped metrics from different partitions would clash and lose their meaning.
2. Being able to write integration tests for multi-partition scenarios. The integration test suite synchronizes with the connector by means of metrics. For instance, there is a typical scenario where the connector waits for a snapshot to be completed. In a multi-partition configuration, the connector should either wait for a specific database snapshotted, or all databases snapshotted. It's impossible to implement on top of the currently existing metrics.