In WaitableWorker, when there are fewer than four jobs to process we
simply process them without going through Sidekiq. When there are more
jobs, we run them in Sidekiq but add a second argument for a JobWaiter
key, and wait for `timeout` seconds to see if those jobs finished.

When there are very many jobs in the batch, we will never finish the
batch before the timeout. This has these consequences:

1. The HTTP request takes longer as we're waiting for the timeout.
2. The jobs can't be deduplicated as the second argument is a unique
   JobWaiter key.

This change adds a feature flag, skip_job_waiter_for_large_batches, that
doesn't use the JobWaiter key when the batch size is more than ten times
the timeout, as we can process approximately ten jobs per second. By
default the flag is off.