ENH: F90 bind(c) intrinsic derived types

[HaoZeke]

This PR will establish the following:

• Intrinsic modules do not cause crackfortran failures
  • Currently only iso_c_binding generates correct wrappers
• The mapping for ISO_C_BINDING types

Additionally, for derived types:

• Scalar numeric types (c_float, c_double)
• Derived types containing arrays of numeric data
• Arrays of derived types

These are being ported over from https://github.com/HaoZeke/f2py_skel/pull/17/files.

The F90 derived type support is in line with #14938.

The implementation is fairly modular, and modifies the global dictionary to generate bind(c) compatible code.

A tracking issue should cover the rest of the items (strings, pointers) which are needed to fully support the standard interoperable C structures.

Requires #15844.

[HaoZeke]

I'm thinking of splitting this up. It is already rather a lot to review. The remaining bullet points have been shifted into #21160.

It would be good to have this in before finalizing the 3-part NEP on F2PY.

@melissawm and @pearu if the implementation and existing tests seem good, then perhaps we can iterate on this in successive PRs?

[HaoZeke]

I understand that without the array support this is more skeletal than it ought to be, but it might still cover enough ground to be useful. If I can't find more time to add the array support (haven't looked it over in a while) it might be useful to just merge it (after rebases of course).

[HaoZeke]

A better / alternate implementation is now almost ready here: https://github.com/HaoZeke/f2py_derived_nep/pull/5/files

One thing to note is that using the PyCapslue method for types which are beyond iso_c seem to require a copy of the input values (so far at any rate).

Needs support from the parser too.