server: Run recover in batches

[future-highway]

The intent of the recovery loop seems to be to recover up to the RECOVERY_LIMIT in increments of BATCH_SIZE, but the limit arg is currently set to RECOVERY_LIMIT.

There also seems to be a typo in how the loop breaks, which could cause an excessive number of messages to be recovered: num_done > RECOVERY_LIMIT instead of num_done >= RECOVERY_LIMIT (equal should also break the loop).

Motivation

This change should result in a more reasonable load per request on external resources (i.e., db & queue) as seemingly originally intended. However, it will likely make the recovery process slower as the number of messages to recover approaches the limit; what was 1 loop will now be up to 100, at the limit.

Solution

Set limit to BATCH_SIZE and stop when num_done >= RECOVERY_LIMIT`.

[tasn]

Thank you for the PR, I asked the team to review!

[svix-gabriel]

Thanks for fixing this!