PEP 688 buffer methods

PEP 688 defines new methods __buffer__ and __release_buffer__ with signatures as follows:

    def __buffer__(self, flags: int, /) -> memoryview: ...
    def __release_buffer__(self, buffer: memoryview, /) -> None: ...

At the moment we have __getbuffer__ and __releasebuffer__ which have slightly different signatures, because they deal in the ffi::Py_buffer type.

It would be nice if we could support the PEP 688 methods, ideally with a way to use the full ffi::Py_buffer flexibility when users need it.

One option could be that we only support the PEP 688 way as first-class, and then allow the user to annotate a method with #[pyo3(unsafe_type_slot = bf_getbuffer)] to directly insert a method into the buffer slot (this could be generally useful for all type slots without requiring us to write any special machinery).

Start by reading PEP 688 and the existing getbuffer and releasebuffer support, then compare their ffi::Py_buffer handling with the proposed Python-level methods. The issue is done when PyO3 supports buffer and release_buffer while retaining a way to use full ffi::Py_buffer flexibility; the exact design remains open.