sz basis smooths can have n factors and the terms for the smooth can be in any order

[gavinsimpson]

Because of a misreading of the {mgcv} help for the sz basis, {gratia}'s current handling of sz basis smooths assumes:

1. that there is a single factor, and
2. that the factor is the first term

(Actually I knew about 1. being wrong but I wanted to get the simplest option implemented first.)

Putting the factor first is simply annoying when fitting or comparing other models with group-level smooths as they all want the continuous covariate(s) first (think fs smooths, but also this is how factor-by smooths work too). And {mgcv} doesn't actually require the factor to be first.

Fix problem 2. and, while doing so, allow for gratia to handle 1. also.

• allow the first covariate mentioned in the smooth to be the continuous term
• allow for more than one factor