Bug #82166 for Marpa-R2: Problem in g->rule_rank_set(5, -2)

I made a fix and uploaded it to CPAN as Marpa-R2-2.033_002.tar.gz. There's a new test for negative ranks in the test suite: rank.t. At the Marpa::R2 level things are unchanged, except that negative ranks now work. At the Marpa::R2::Thin level, I've changed the calls to check for whether an error actually occurred before throwing one. A -2 return is ambiguous (that is, it can be either success for a rank of -2 or failure). This is now documented. At the Libmarpa level, I cleaned up the code, and made sure that the calls which return and/or set ranks always clear the error code back to MARPA_ERR_NONE on success. This allows the upper levels a safe way to clearly determine whether or not a failure occurred. I've revised and clarified the Libmarpa documentation. Interfaces whose return codes do not indicate errors unambiguously, and which need to be double-checked, are known. They occur in cases like this, where "natural" return values cover most of all the possible return values, leaving no obvious choice of a return value to use for exceptions and/or errors. But they're unusual -- I don't really like them and I imagine my sentiment is shared. Most libraries figure out some way to avoid ambiguous error returns. But this interface preserves backward compatibility --any code which currently works will continue to work.