Strange alignment result

While trying to write a test for issue #299, I discovered strange behaviour in the Gotoh alignment algorithm that we're using. Here's a small script to reproduce the problem:

import gotoh

gap_open = 40
gap_extend = 10
use_terminal_gap_penalty = 1
reference = 'EFACDE'
query =     'SFACDS'
aligned_ref, aligned_query, score = gotoh.align_it_aa(
    reference,
    query,
    gap_open,
    gap_extend,
    use_terminal_gap_penalty)
print(aligned_ref)
print(aligned_query)

The results are:

-EFACDE
S-FACDS

If I replace the E's with H's, I get this:

HFACDH
SFACDS

I understand that some substitutions are more likely than others, but why would it behave differently at the start and the end? If gap_open + gap_extend <= 10, then it opens a gap at both ends for E, and if it's less than or equal to 5, then it opens a gap for H.

Start by running the supplied reproduction script and inspect the implementation behind gotoh.align_it_aa, focusing on terminal gap handling and the substitution scoring that produces different results for E and H. Compare the alignment and score in both cases; done means the asymmetric behavior is explained and, if incorrect, corrected with regression coverage.