Use prometheus to annotate pod/node data onto Job model

[jjnesbitt]

Supersedes #697

This PR collects much more data every time job data is collected by the job webhook. This allows us to derive "cost per job" metrics.

The following notable changes are included in this PR:

• build_timings_processor.py has been moved/renamed to the job_processor folder, with several files, including one for build timings. This is because at this point, a lot more is going on than just uploading build timings.
• A new Node model, which stores data relating to a node that a job ran on, including available cpu/memory, its spot price, etc.
• A new JobPod model, which stores data about the pod a job ran on in the cluster, including resource requests, limits, and usage.

The above two models are referenced by Job, but are null-able, since jobs that weren't run in the cluster (uo, etc.), as well as historical aws jobs, won't have any available data.

There are a few more minor changes worth mentioning:

• The Job.duration field has been changed from a FloatField to a DurationField, to better represent the data and make it easier to work with.
• The Job.aws field has been changed to be no longer null-able. The field was made null-able in Handle jobs with missing runners #724, to get around a quirk of the gitlab API. This PR circumvents that part of the API entirely, and so null values shouldn't be allowed. The amount of rows in the database that have a null value in the aws column is only a handful at the moment, and so those rows are deleted as a part of the migration, to clean things up.

[jjnesbitt]

@alecbcs @cmelone Let me know if anything seems awry or jumps out at you here.

[jjnesbitt]

I've been testing this and am seeing times of around 3-5 seconds for the process_job task.