Feature/one splits file to rule them all

[hlgp]

Merge NonShardedSplitsFile and ShardedTableMapFile. Use TableSplitsCache instead of reaching out to Accumulo in every reducer.

Splits File generation now takes configured Partitioner needs into account.

• If the table's partitioner doesn't need splits, e.g. HashPartitioner, don't write them
• If the table's partitioner only needs splits, only write the splits
• If the table's partitioner needs splits and locations, write them
• If the table's partitioner needs splits in a specific order, write them in that order, so we do not need to sort them at the beginning of every mapper.

[austin007008]

While you're in here you might want to fix the test data in this and the corresponding unit tests. those "splits" are actually memory locations from something (I can't remember what) and they should be some kind of base64. The only reason they don't fail on the encode/decode is because apache commons base64 will happily let you input and output bad data . you can see https://github.com/NationalSecurityAgency/datawave/pull/2480/files for an example of how they should look