Essential changes:

• lift the limitation for 384-bit modular inversion, which doesn't benefit BLS12-381 applications, but future-proofs blst_t.hpp, a generic interface to assembly modules
• harden ptype##s_mult_wbits
• tolerate infinity points even in ptype##s_precompute_wbits
• smooth some edges in bindings/blst.hpp
• add sub_aggregate method to Rust bindings