Clean up packaging of conversion methods

[fnothaft]

Continued from #1157. We should:

• Do a review pass and make some of the AlignmentRecordConverter methods private
• Make SAMFileHeaderWritable private to ADAM, and change the AlignmentRecordConverter.convert signature to take a SAMFileHeader

Also, contributed for discussion by @heuermh:

Also, I think it's more useful to have the conversion utility at the record level, rather than the RDD level, since then it can be used outside Spark.

If that were the case, it would be preferable to me to have the conversion stuff written in Java instead of Scala and be a library external to ADAM.

[heuermh]

A proposal for said external java library here:
https://github.com/heuermh/bdg-convert

It is inspired by Commons Convert but uses Guice instead of a custom registry. Then I like the concept of htsjdk's ValidationStringency but don't like the name. Please propose something better (and shorter) than ConversionStringency.

[heuermh]

I've started fleshing out conversions from Biojava and my own dishevelled library, here in a separate repo due to LGPL licensing of the dependencies:
https://github.com/heuermh/dishevelled-bio/tree/master/convert

It also depends on the bdg-formats Sequence/Slice/Read pull request, so won't build without the right snapshot dependency.

I think there is enough in the bdg-convert repo and here to demonstrate the proposal.

[heuermh]

Split the external library proposal into a separate issue #1197 for further discussion. Let's keep this issue focused on the two visibility items mentioned above.

[fnothaft]

I'm triaging this to 0.24.0.