At the moment, we enqueue modules to the task queue via these two endpoints:

• /poll-and-queue: enqueues new modules from the index

• /requeue: enqueues modules with status 50x

As a result, when we are are reprocessing modules, new modules might end up taking longer to be proccessed. For example: #39564 and https://gophers.slack.com/archives/C0VPK4Z5E/p1591981603268600.

Rather than calling s.queue.ScheduleFetch in handleIndexAndQueue, we should change the enqueuing logic as follows:

• Separate /poll-and-queue and /requeue into these two endpoints:

• /poll: polls the module index and inserts into module version states

• /enqueue: call postgres.GetNextModulesToFetch to figure out which modules to enqueue next

• Change postgres.GetNextModulesToFetch to prioritize enqueuing modules with status=0, ahead of modules with status 50x

• Deprecate /poll-and-queue and /requeue