Support IDBlock's update in CAS mode when use HBase backend

**ConsistentKeyIDAuthority#getIDBlock** are not so efficient, it works like this(Suppose we use HBase backend)
1. make a Get request to obtain all the Cells of the target rowkey(value is partitionKey)
2. Iterator the returned Cells to determine the _nextStart_ ID, and compute the _nextEnd_ ID (_nextEnd_ = _nextStart_ + blockSize)
3. make a Mutate request to write a target Cell, it’s column name start with _nextEnd_ (ConsistentKeyIDAuthority#getBlockApplication)
4. make a Get request to obtain all the Cells which column name prefix with _nextEnd_, If our claim is the lexicographically first one, we own this id block. Else we should delete the Cell and try again.

After a time, there can be lots of Cells with the target partitionKey, and the query efficiency will be very low.
So how about do this in CAS model, since hbase already have **Table#checkAndPut** to support this. And just one Cell with each partitionKey is enough.

Start by tracing ConsistentKeyIDAuthority#getIDBlock and getBlockApplication to understand the current HBase reads, mutations, and claim checks. Review HBase Table#checkAndPut and determine how the CAS flow would preserve ID-block ownership with one Cell per partitionKey; done means the proposed update works correctly in CAS mode without accumulating competing Cells.

Written by the indexing model from the issue text.