Add "deployment" lines to Honeycomb traces for before/after comparison

[lucyb]

A recent incident with slow performing queries was caused by a regression in cohort extractor. It took a while to track down. In future, it would be helpful to be able to associate code changes with job performance/behaviour. Sending details of deployments into Honeycomb would give us this information.

[madwort]

Use markers.

For job-runner changes & cohort-extractor changes.
Problem: we don't have a formalised way of saying when a deploy happens.
Would be easier if we had docker-ised job-runner.