Bf udf

[kortemik]

spark allows running multiple aggregations in the same run by using

      .agg(tokenAggregatorColumn0, tokenAggregatorColumn1, tokenAggregatorColumn2)

therefore simplifying this so that:

• Tokenizer is now it's separate UDF, run this before applying the BloomFilterAggregator
• BloomFilterAggregator has no longer overhead for it's internal byte[] (de-)serialization between rows

one gets the separate bloomfilters to their dedicated columns, use BloomFilter.readFrom(bais) and .expectedFpp to check if the filter is healthy and should be used. healthy can be checked by comparing it to set fpp (false positive propability).