Also cleanup any related code, like the use of extractFromSelf.

After the change to make each batch contains all of its own measurements, moving away from the old approach of having different measurements writing to the same bucket sharing a shared_pointer to the write batch, the de-duplication work inside of determineBatchesToCommit is no longer necessary. We still need to sort the batches by bucket order to prevent deadlocks (two clients trying to commit writes to buckets in different orders atomically).

Besides de-templating the function, we can also repurpose it to only perform the sorting functionality.