Make uid range partitioning fast

[EnricoMi]

Uid Partitioning uses Dgraph's GraphQl operation offset, which is very slow on large result sets (see issue #5808). The alternative after has constant response time. Remove the offset related pagination used for partitioning and use the chunked reading that is based on after.

This removes the count uid range estimator (because count is also slow, see issue #5813) and any offset related code from PartitionQuery.

If at some point count and offset become efficient, we can resurrect this code.