Add partition awareness to base task components

[morozov]

In #15, there's the TaskOffsetContext implemented that contains all partitions and offsets to be processed by a task. Next, we need to implement the iteration over all task partitions in the event coordinator by introducing a parallel "partitioned" event source class hierarchy. There are two points to this:

1. Not all connectors are partitioned by design (the MySQL one isn't).
2. By introducing a new API, we'll be able to continue with the changes only for SQL Server and simplify the migration of other connectors to the new API later.

Acceptance criteria:

1. Introduce the following classes and interfaces:
   • PartitionedChangeEventSourceCoordinator
   • PartitionedChangeEventSourceFactory
   • PartitionedSnapshotChangeEventSource
   • PartitionedStreamingChangeEventSource
     Apart from being parametrized with <O extends OffsetContext>, these APIs should be parametrized with <P extends TaskPartition>. This will allow the underlying components to use the database name of the currently processed partition via partition.getDatabaseName(). Copy/paste where extension is not possible (e.g. the coordinator).
2. Alongside with the OffsetContext offsetContext arguments introduced in DBZ-2975: Extract offset context from object states to method signatures #13, add the TaskPartition partition argument where necessary.
3. Move the loop introduced in DBZ-2975: Prototype multiple offset contexts per task #15 to the partitioned coordinator: https://github.com/sugarcrm/debezium/blob/efc7d20e4d973ee5e04863b5923f154ae41a7c38/debezium-connector-sqlserver/src/main/java/io/debezium/connector/sqlserver/SqlServerConnectorTask.java#L128-L132 Leave the break temporarily, repeat for each source (snapshot, streaming).
4. Update the SQL Server connector components to implement the partitioned API.

UPD: work on this issue showed that there's more classes apart from the Coordinator which need their APIs changed. Copy-pasting more code will increase the likelihood of being not in-sync with upstream when the changes are made to the duplicated code. In order to expedite development and lower this risk, let's take a shortcut:

1. Instead of introducing new class hierarchy, add partition awareness to the existing one.
2. Temporarily delete all plugins except for the one for SQL Server that would otherwise require code/API changes.

[mikekamornikov]

Closed via #22 .