Buffer object additions

If you make something a buffer type, this adds the ability to do:

ob_vals = memoryview(ob)
# or
ob_vals = np.fromarray(ob)

ob_vals[:] = 0
# etc.

There are a standard set of operations that a buffer object will tend to have. [] indexing, length, etc, and they all can be defined with the structure defining the buffer object. Is there a way to set up these standard helper methods? Something like this:

.def_buffer([](MyArray &r) -> py::buffer_info {
     return py::buffer_info(
         r.data(),                         // Pointer to buffer
         {static_cast<ssize_t>(r.size())}, // Diminsions
         {sizeof(float)}                   // Strides (in bytes) for each index
         );
    },
   py::buffer_helpers::getitem<1>(), // 1D
   py::buffer_helpers::setitem<1>() // Etc
)

These helpers could add things like (based on test_sequences_and_iterators):

• __getitem__, with slicing support, following numpy for 2+ D
• __setitem__, as above
• __len__
• __iter__

Some of these are a bit tricky to add manually, and are fully defined by the buffer_info struct.

Is something like this practical in pybind11? The operators py::self seem to be able to do something like this, and py::pickle, but I didn't see a more general extensible mechanism for adding named methods. (But, since this needs to happen for .def_buffer, it might make sense to add something just for this).

As a side note, the shorter, cleaner, less general versions of py::buffer_info are not mentioned in the documentation, though they are used in the tests.

Start with the .def_buffer entry point and the test_sequences_and_iterators reference in the issue. Determine whether buffer_info can define reusable indexing, assignment, length, and iteration helpers, and review how py::self and py::pickle add named methods. Done should include a settled API direction, coverage for the proposed behavior, and documentation for the shorter py::buffer_info forms.