Bigtable: Implement query sharding by generalizing ReadRows resume request builder.

[igorbernstein2]

This extends the work done in #2986 to allow map reduce style frameworks like beam to split queries into shards and execute them in parallel. The mechanism for chopping off part of query in a resume request is very similar to splitting a query into multiple shards. The main difference is how many splits are used.

The common functionality is extracted to an internal RowSetUtil class that does all of the heavy lifting. The class is used both byReadRowsResumptionStrategy for computing the resume request and the newly introduced Query#shard method.

Also expose the ability to get a Query's bounding range. The combination of Query#shard & Query#getBound is needed to implement a Beam source