rework message queue event model

[vredchenko]

Initial MQ events need to be reworked.

• add missing DB indexes, Add missing indexes for 4 DB tables smartem-devtools#44
• The API now mutates DB directly, bypassing the MQ - but it still needs to alert consumers on the MQ to every DB mutating operation
• attach decision-making logic placeholders to listen to MQ events
  • initialise prediction model weights; inputs are: prediction_model_name and grid_uuid; would be the first thing and happen once per model per grid at the start
  • generate random model predictions; inputs are: prediction_model_name, grid_uuid, level=hole|square; when gridsquares and foilholes are first registered they need initial predictions
  • random prior updates; inputs are: grid_uuid, origin=motion_correction|?; will come from whatever metric we decide to use as feedback. The origin would likely be motion correction, ctf or particle picking, but there are a few options for metrics to send back
  • register quality prediction model; inputs are: model_name, model_description; would only happen as a setup stage, probably by hand but could be automated. I think to avoid complications any existing sessions would not be touched by that
• clean up HTTP API DB_WRITE_MODE - only direct mode remains
• clean up mq_consumer.py
• update docker container (build new version and publish to reflect changes in this PR)
• test that watcher and not just parser works correctly and persists to DB: test EPU agent watcher persists data to DB (not just parser) smartem-devtools#50

[d-j-hatton]

The placeholder logic seems correct to me. The only suggestion is the one above which is something I should have done in the original. It just makes sure that the initial weights are evenly distributed against whatever models are registered and that the weights sum to 1