Found a few more edge cases when applying this in production,
so I'm patching this migration retroactively.

Trying to remove a label that was not associated with an event
gave a 400 HTTP response. Fixed to be 404. It was already correct
for superevents.

Things like 's190418BC' will work now (should be 'S190418bc')

Add a pipeline_type field to the Pipeline model. This will facilitate
grouping of pipelines into different classes.

It's necessary for kubernetes health checks.

Now there are two separate steps - get notifications and
get contacts for those notifications. Makes testing easier.

This is not needed anymore with the changes to the group infrastructure
which add an optional 'ldap_name' to groups whose membership should be
managed by LDAP queries and shib sessions.

These tests cover the bugfix in the previous commit.

Based on "feedback" from the low-latency group, the p_astro
and other information which is in the VOEvent files will not be
made available through the API to unauthenticated users. I'm
not sure why.