Rewrite rules/macros in Starlark for greater flexibility

[nicholasjng]

This project started as a small collection of macros for nanobind in Bazel, heavily borrowing from pybind11-bazel and its macros.

Some design differences and changes from nanobind (most notably, the library part and the stable ABI facilities) have made it necessary to expose a number of feature flags to fully take advantage of nanobind's configurability.

Unfortunately, the combination of different features (SABI versions, target platforms, build mode, now free-threading coming up) have lead to a combinatorial explosion of config options in nanobind-bazel.

I'm not sure how many people inspect the Bazel graph on the daily, but I imagine nanobind-bazel's to look pretty horrifically cluttered.

What would be a better solution is to rewrite the macros as rules with proper Starlark implementations, where you can (if I'm not mistaken) get a flag's value, and proceed to do different things with if statements based on the obtained values.

This would also make it easier to just ignore certain ill-formed combinations, e.g. SABI builds with Python versions <3.12. It would also help to not produce two .so build artifacts in the SABI case, one of which is completely unused.