refactor drop_trends handling

[tsostarics]

currently, warnings about dropped trends only comes up with set_contrasts. Something like enlist_contrasts(df, x ~ sum_code - 3:4) will just ignore the - operator at the level of use_contrasts.

I think this can be refactored entirely at the level of .make_parameters, specifically with .process_code_by by checking if params[['code_by']] is a polynomial scheme and if params[['drop_trends']] is not NA. If both are true, then the latter can be reset to NA and a warning can be issued, which will be present regardless of set/enlist/glimpse_contrasts. .reinstate_dropped_trends can then be removed entirely, omitting the regular expression wellformedness check.

[tsostarics]

just a note but most of this can be put on .make_parameters, but there does need to be some way to check whether this is occurring within set_contrasts or not, otherwise set_contrasts(df, x ~ contr.poly - 3:5) will still work when it shouldn't

[tsostarics]

This is implemented now, and .reinstate_dropped_trends has indeed been completely removed 🎉.

The implementation has set_contrasts set an attribute on the list of formulas, which is checked in enlist_contrasts. If present, .process_contrasts will ignore the - operator via the omit_drop parameter. The user can replicate this behavior by setting the omit_drop attribute themselves but by and large i doubt people would bother.

.make_parameters will handle the separate warning when the - operator is used with non-polynomial scheme functions. This already prevents the drop_trends parameter from being set to anything besides NA, so you don't get 2 warnings for things like set_contrasts(mtcars, carb ~ contr.sum - 4:6).

The one downside is that the user will get multiple warnings for every formula containing a - operator when used with a polynomial scheme in set_contrasts rather than a single warning listing all the problematic formulas. But, hopefully this drives home the point that this usage is not supported.