Stop pickling RecordTransmitters and serialize its configuration instead

[jondequinor]

• A storage session in a remote environment (e.g. a unix prefect task), assumes to find storage_server.json in order to know how to connect to storage. A RecordTransmitter should not rely on storage_server.json, it should inherently know how to connect.

• RecordTransmitter is serialized using pickle. This is probably not a good idea for a IO-bound instance, as the same instance will likely have to operate in two wildly different environments (e.g. RGS vs docker in azure). The configuration for a RecordTransmitter should be serialized instead. Also, pickle will not call __init__, which might cause confusion in the long run.

Definition of done

1. RecordTransmitters are no longer pickled
2. RecordTransmitters inherently know how to connect

[jondequinor]

An implementation was started here https://github.com/equinor/ert/pull/2050 but abandoned because it did not fix the issue it was trying to fix.

[eivindjahren]

Closing as related to ert3, which is no longer the direction taken by the project

[sondreso]

This is part of ert.data, and is still relevant

[eivindjahren]

There are is longer any class RecordTransmitter in ert.data so there is no pickling of this class. I am closing the issue. If there is some other class that should not be pickled, we can make a new issue for it.