Remove the non-existent extended precision scalar types

Remove the following `np.number` subtypes, and the related `npt.NBitBase` stuff.

- `int128` / `uint128`
- `int256` / `uint256`
- `float80` / `complex160`
- `float256` / `complex512`

This way, the only remaining extended precision types are

- `float96` / `complex192`
- `float128` / `complex256`

It's probably a good idea to also clean up `_typing._extended_precision`.