Implement shortnull redux_gl for exon-level

[lgragert]

Here's the current py-ard redux_gl for A*01:04N.

A*01:04N   lg   A*01:01g
A*01:04N   lgx   A*01:01
A*01:04N   G   A*01:01:01G
A*01:04N   exon   A*01:04:01
A*01:04N   W   A*01:04:01:01N/A*01:04:01:02N

Based on the shortnull rule for propagation of expression characters, A*01:04N should map to A*01:04:01N since both four-field alleles have the same expression character "N".

To compute two-field population freqs (w/o ARD rollup), I had to truncate the exon-level three-field GL strings to two-field, but some of the three-field GL strings are incorrect because of this issue.

WHO (W) mode is correct.