I forgot that Object doesn't include nil on protocols.
I added test coverage demonstrating the regression, only to note that
parameterized and literal queries produce different sql for the nil
case. I believe only the former is actually valid ANSI SQL, though mysql
at least does not care.
This allows seqs to be used as param values, e.g. for IN clauses. It
also converts sets to seqs for the same purpose.
It does not expand the parameter name when this occurs, which will lead
to a mismatch between the *param-names* and *params* collections. It is
not clear how *param-names* are intended to be used, so I have no better
fix to suggest other than repeating a seq's param name the corresponding
number of times into that collection.
