[mypyc] feat: true_dict_rprimitive for optimized dict fastpath usage

[BobTheBuidler]

This draft PR adds a true_dict_rprimitive which allows mypyc to differentiate between an actual dict instance and a subclass of dict.

Since this impacts any function that builds kwargs, plus dict comprehensions and other things, the special casing seems reasonable.

I chose not to map builtins.dict to the new rprimitive because we cannot guarantee types of input arguments. I intended to ONLY use it in instances where the dict type is certain. I'm not sure if this decision is correct or not, looking for opinions.

There is one test failure I still need to figure out, but I'd appreciate any feedback on the direction.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[brianschubert]

(Just a passing remark, I haven't looked at the actual change)

I think "exact dict" would be a better term than "true dict" to describe instances of dict not including instances of subtypes of dict. Similar terminology is used in Python's C API, e.g. PyDict_CheckExact.

[BobTheBuidler]

Should we maybe just rename dict_rprimitive to dict_subclass_rprimitive?

I noticed while coding this PR that it isn't mapped the same way as the other rprimitive types, and this is not clear from the definition in rtypes, so maybe renaming that var and then changing true_dict_rprimitive to dict_rprimitive is the optimal choice.

[BobTheBuidler]

I decided to go ahead and map builtins.dict to the new primitive, and updated the naming as per @brianschubert recommendation. Everything seems to be working except for some issue with defaultdict. For some reason, defaultdict is now also mapped to the new dict primitive. but my if check is very clearly name== "builtins.dict" which should not be the case for defaultdict?

If I remove my new line from the mapper, which leaves "builtins.dict" mapped to the existing primitive, all tests turn green.

If we could decide we want the check, we need to figure out why mypy is treating defaultdict like a non-subclass dict. One of you guys might already know why. Maybe mypy handles defaultdict uniquely. It might be the case that this is not possible to workaround in a reasonable fashion.

I need an opinion from a maintainer on the best way to proceed here. I could remove the new check from the mapper and we could merge this in basically now.