Clean up alignment record materialization logic

[fnothaft]

I was using mango to debug some variant calling issues and ran into a few nits in the alignment backend:

• Using Java IO libs instead of Hadoop IO libs to access index files
• Inconsistent file format checks between VizReads and AlignmentRecordMaterialization
• It's valid for BAI's to be at either <filename>.bam.bai or <filename>.bai so we should support that
• We should also add support for CRAM

@akmorrow13 I'll follow up with you tomorrow; I can take these on later this week but had a dev environment issue that I needed to run by you first.

[akmorrow13]

@fnothaft Great, I've been meaning to address most of these. We can talk specifically about points 1 (java IO) and 4 (CRAM support) tomorrow or after Tuesday.