Fix TypeError for boolean opt-out fields when optout_col_values contains non-boolean values

[martinburchell]

Fixes #140 so that for boolean opt-out fields, non-boolean values are dropped from optout_col_values in the anonymisation config file.

@RudolfCardinal A side-effect of this is that if there are no valid values for optout_col_values, all patient ids will be considered as opted out. Is this OK?

In adding some tests, I've also added support for database-level testing from pytest, initially with SQLite and MySQL. It may well work if you pass a non-MySQL url to pytest but I've only tested with MySQL. This will change when I add this functionality to #141 and we support more backends.

I've also changed the names of some other test classes, which were previously being ignored by pytest.

[martinburchell]

@RudolfCardinal as discussed, the script now aborts with ValueError (as with the other configuration checks) if there are no valid values for optout_col_values. I suggest we merge in order #143, #141 before this one (with any fixups) once they have your approval.

[RudolfCardinal]

Looks good! I did wonder about the part in validate_optouts() in which the check for completely absent values in optout_col_values only occurs if the column is boolean (which is when it may just have been trimmed). I thought: should that check apply regardless of column type? But then I thought: this table might be a list of opt-out PIDs (e.g. a table with a single INT column), in which case this may be absolutely correct -- you don't care what the opt-out value is, because anything counts as an opt-out, and that's what you code correctly does. So I think spot on as it is.