Implement distributable real-time gain calibration

Agreed: Towards distributed code: Receive visibilities in a distributed fashion      Demonstration: LEAPacc to ingest a mock datastream Stream partial solutions       Demonstration: LEAPacc writes it's solutions to a data queue Stretch: Implement stream processors that aggregate     Demonstration: A aggregator function that can receive the output from the distrubuted LEAPacc (for interpolate for TAB directions in future PIs) Previous: Pipeline script available that can: Receive visibilities in a distributed fashion using the real-time visibility streaming interface (i.e. ska-sdp-realtime-receive-processors / ska-sdp-dal), i.e. different processes are configured to receive different frequency chunks (as configured in an execution block). Stream partial solutions / normal equations to data queues (Kafka topics) using the existing libraries (ska-sdp-dataqueues) This might involve defining a suitably flexible xarray-based data model for partial solutions / normal equations Stretch: Demonstrate that this can be distributed and scaled by streaming data to different Kafka topic partitions Implement stream processors that aggregate and solve partial solutions / normal equations and emit the (possibly also partial) solutions back on Kafka queues Stretch: Allow re-visiting visibility data for refining solutions at later stages - likely somewhat tricky as it would require keeping visibilities in-memory Implement final aggregation / interpolation step that generates solutions for CBF onto the queues Stretch: Use previous solutions to seed / linearise subsequent distributed real-time calibration solver iterations. Might require changing normal equations data model? Stretch: Demonstrate integration with receive (allow working on live SPEAD stream) Stretch: Demonstrate integration with queue connector (expose calibration solution over Tango)