💥 concrete floating scalar types

[jorenham]

backwards incompatible bugfixes:

• float96 and float128 are now (equivalent) aliases of longdouble
• longdouble.item() and .tolist() now correctly returns longdouble instead of builtins.float
• float16, float32, and longdouble have been made into proper floating subtypes
• moved the is_integer and as_integer_ratio definitions from floating to its concrete subtypes

---

follow-up of #391
closes #198
towards #136