This is the solution that @yfarjoun and I have agreed on. In the presence of
CIGAR with >65535 operators, we move the real CIGAR to the CG tag and put a
fake full clipping CIGAR (i.e. <readLength>S) in place.

`./gradlew` reported one test failure due to `EnaRefServiceTest`. Looking at
the call stack, I don't think that is my fault. The rest of tests have passed.

On an example SAM and BAM (from <http://lh3lh3.users.sourceforge.net/data>),
`PrintReadsExample` gives the desired BAM output when taking both SAM and BAM
as input. Both shallow and deep decoding also write correct BAMs.

For BAMs generated with old tools, the indexingBin field may not be correctly
set. We need to manually update this field to avoid errors during random
retrieval.

The old code to lift tag to CIGAR did not work. This commit fixed several
issues:

- byte order is not set when computing alignmentEnd
- forgot to add *4 to cigarLen in two cases
- getChar() is not what it is supposed to mean

The new code has been more carefully tested and seems to work.

This fixes a test failure.

- extracting constants etc.

…ed the SamSetRecordBuilder..

…ed the SamSetRecordBuilder..