BQSR: avoid throwing an error when read group is missing in the recal table, and some refactoring.

[takutosato]

Addresses #6242.

Current behavior: when all the reads in a read group are filtered in the base recalibration step, the read group is not logged in the recal table. Then ApplyBQSR encounters these reads, can't find the read group in the recal table, and throws an error.

New behavior: if --allow-read-group flag is set to true, then ApplyBQSR outputs the original quantities (after quantizing).

I avoided the alternative approach of collapsing (marginalizing) across the read groups, mostly because it would require a complete overhaul of the code. I also think that using recal data from other read groups might not be a good idea. In any case, using OQ should be good enough; I assume that these "missing" read groups are low enough quality to be filtered out and are likely to be thrown out by downstream tools.

I also refactored the BQSR code, mostly to update the variable and class names to be more accurate and descriptive. For instance:

ReadCovariates.java -> PerReadCovariateMatrix.java
EstimatedQReported -> ReportedQuality