rationalize Index.select(...) return/exception values

[ctb]

In #1936, I realized that when confronted with a select that will result in an empty Index class, some classes (LCA_Database in particular) raise a ValueError, while others just return an empty Index class.

See test_index_select_nada in particular.

I'm not sure what it should do but it would be good to have a single kind of behavior. I think I originally implemented it this way so as to provide good error messages to people when loading the wrong k-mer size for an LCA or SBT database.

[ctb]

I'm not sure what it should do but it would be good to have a single kind of behavior. I think I originally implemented it this way so as to provide good error messages to people when loading the wrong k-mer size for an LCA or SBT database.

I tried changing LCAs and SBTs over to returning an empty database on incompatible select in #2204, but I then reverted that change, because so much useful information was lost in the process!

Instead, once #2204 is merged, we just make an empty database in sourmash_args.load_dbs_and_sigs when select raises a ValueError. This leaves the internal API a little messier, but keeps informative error messages around.

[ctb]

#2204 will close this with #wontfix.