don't classify desc as romanization if it is in data["categories"]

[jmviz]

Noticed some spurious romanizations. Consider for example Reconstruction:Latin/sufferio. parse_word_head() was going through the cleaned version of {{la-verb|4|*sufferiō|*sufferīv|*suffert}} {{lb|la|Proto-Romance}}, which is *sufferiō (present infinitive *sufferīre, perfect active *sufferīvī, supine *suffertum); fourth conjugation (Proto-Romance), and was determining Proto-Romance to be a romanization. This can be ruled out as a romanization since {{lb}} generates the category Proto-Romance, which is in data["categories"]. So this PR adds a check to make sure cases like this are avoided.

Now, the spurious romanization is gone, but there now is a DEBUG: unrecognized head form: Proto-Romance that gets logged as the desc ends up not being classified. I left this alone as I'm not sure what the general system is for these classifications. (It's possible this check should be moved up higher in the for desc_i, desc in enumerate(new_desc): loop, depending what that system is...)

(There happens to be an unrelated Lua execution error on this page when invoking {{VL-conj-4th}}, but that is neither here nor there for this PR.)

[kristian-clausal]

This seems like a sensible change; a situation where a romanization and a category are identical and present on the same exact page (especially when a vast majority of categories are at least a couple of words) should be extremely rare or even non-existent.

[kristian-clausal]

After testing out changing things to skip the debug message (by inserting the category check inside the romanization branch which would have continue'd the loop and bypassed the debug message), I actually realized the correct way to fix this (particular example) is to add Proto-Romance in uppercase_tags in tags.py, so that Proto-Romance gets picked up as a tag. That way, it doesn't end up in romanizations and stops the debug message from triggering.

But that doesn't mean the code here is wrong in general; it is still true that if we see a romanization that looks like a category from that page it shouldn't be used as a romanization. In that case, the debug message is useful information.

[kristian-clausal]

Also was able to figure out the Lua thing:
tatuylonen/wikitextprocessor@0543061
tatuylonen/wikitextprocessor#35

It was an issue with super-weird code in Module:VL-verb; the __tostring metamethod was used to pass data in the form of tables (instead of strings), which is a thing that was not prohibited by 5.1, but has been prohibited in later versions of tostring() (the function calling __tostring).