Improve dtype for compound datasets

[loichuder]

Let's take the example from Numpy:

with h5py.File("test.h5", "w") as h5file:
    h5file.create_dataset(
                "dogs",
                data=np.array(
                    [("Rex", 9, 81.0), ("Fido", 3, 27.0)],
                    dtype=[("name", "S10"), ("age", "i4"), ("weight", "f4")],
                ),
            )

When asking for the metadata, h5grove was returning:

{
  ...
  dtype='|V18'
}

Now it returns

{
  ...
  dtype={"name": "|S10", "age": "<i4", "weight": "<f4"}
}

[t20100]

It seems compound types are recursive...

with h5py.File("test.h5", "w") as h5file: 
    h5file.create_dataset( 
        "dogs", 
        data=np.array( 
            [("Rex", (9, 81.0)), ("Fido", (3, 27.0))], 
            dtype=[("name", "S10"), ("combo", [("age", "i4"), ("weight", "f4")])], 
        ), 
    ) 

Do you want to support that?... else that would give {'name=': '|S10', 'combo': '|V8'}

[t20100]

LGTM, unless you want to handle recursive compound types.

[loichuder]

Do you want to support that?

Yeah it is not much to add to support it. It just make the return type recursive which cannot be linted by mypy as it is not supported 😞 python/mypy#731