Improve performance of task claims by managers

[bennybp]

Description

We've noticed that with large databases, the time to claim tasks is a bit slow (multiple seconds). This is particularly noticeable in Issue #850 .

This PR aims to improve this significantly via two main ways:

• Improving the SQL query. Not the task queue has an available column which can be used to filter, rather than the status column of the joined base_record table. This seems to be the fastest way, although it denormalizes stuff a bit
  • I tried a conditional WHERE status = 'waiting' index on base_record but that didn't even help as much
• Improving the task specification generation (which happens sometimes when claiming) so that data is retrieved in fewer SQL queries. Also, I eliminated creation of pydantic objects here, which also trims a bit of time.

Tests on a copy of a production database show a reduction from ~2-3 seconds to 0.5-0.7 seconds for 200 tasks. Getting it much lower is going to be tough, since some of that is data transfer overhead over the internet.

Note that this does require a database migration with the upgrade-db subcommand.

Status

• Code base linted
• Ready to go

[bennybp]

Not seeing any problems with this on my testing instance, so looks good to me