Need to discuss with Nick on multiple messages vs single messages with both results

@ndevenish says it will be multiple messages. For implementing this lets assume the format is:

{
    "results": [
        {
            "centre_of_mass": [
                2.207133058984911,
                1.4175240054869684,
                13.317215363511659,
            ],
            "max_voxel": [2, 1, 13],
            "max_count": 702.0,
            "n_voxels": 12,
            "total_count": 5832.0,
            "bounding_box": [[1, 0, 12], [4, 3, 15]],
        }
    ],
    "status": "success",
    "type": "3d",
    "source": "gpu",
}

It'll be close to that I suspect. Nick, feel free to correct us if you have a better idea of the format.

You are listening to messages directly, right? Are you getting the messages back from the xrc.{ispyb_beamline} queue?

It might be in the message parameters, along with dcid/dcgid, is that okay?

As discussed with @DominicOram:

• We want a parameter to turn on GPU processing and a separate one for using the GPU results
• If use_gpu_results is on, then always use the first message we recieve from ZocaloResults. Otherwise, use the CPU results

As part of this issue: if we are using gpu results, we would like to compare data from GPU and CPU results and include it in a debug log, and warn if the results differ by some tolerance (TBC)

This should be done asynchronously so we aren't still being blocked by slow CPU results

Updated acceptance criteria:

• Add parameter to use zocalo gpu results
• If this is true, then wait for zocalo results from both cpu and gpu. Warn if the CPU results arrived first, compare results from CPU and GPU, and warn if they are different at all (within a floating point rounding error).
• If waiting for zocalo results timeout and we only received one set of results, then warn and use those results
• If use zocalo_gpu_results is false then just use CPU results as before
• Create follow up ticket: After we're confident that GPU results are correct, change behaviour so that we no longer wait for both results to arrive and instead only wait for the first result

Changes are currently on the beamline to use the CPU results no matter what. Update this accordingly