ENH: do not explicitly import spm._spm

[balbasty]

See #3

This PR removes the explicit import of spm._spm and instead lets the wrapped package (in our case, spm-python) implement its own Runtime class. Something like:

# in: spm/__wrapper__.py
from mpython.runtime import Runtime as RuntimeBase

class Runtime(RuntimeBase):
    @classmethod
    def _import_runtime(cls):
        import spm_runtime
        return spm_runtime

It also needs each of the MatlabClass children it implements to point to the correct runtime. It's a bit tricky if we don't want to mess with the __init_subclass__ business implemented in MatlabClass but I think it can be done in a not too verbose way using a mixin:

# in: spm/__wrapper__.py

class RuntimeMixin:
    @classmethod
    def _runtime(cls):
        return Runtime

# in: spm/nifti.py
from mpython.matlab_class import MatlabClass

class nifti(RuntimeMixin, MatlabClass):
    ...

I am leaving this as draft as it needs

• spm-runtime to be properly released and published to pypi
• spm-python and mpython to define the runtime and mixin described above

---

I've also replaced the local imports (used to break circular imports) with "delayed imports" using a magic class. The (only?) advantage is it allows to have all the imports listed at the top of the file like we're used to. See e.g.

mpython-core/mpython/core/base_types.py

Lines 8 to 15 in ec0f3e5

	class _imports(DelayedImport):
	Array = 'mpython.array.Array'
	Cell = 'mpython.cell.Cell'
	Struct = 'mpython.struct.Struct'
	SparseArray = 'mpython.sparse_array.SparseArray'
	MatlabClass = 'mpython.matlab_class.MatlabClass'
	MatlabFunction = 'mpython.matlab_function.MatlabFunction'
	AnyDelayedArray = 'mpython.core.delayed_types.MatlabFunction'

---

Happy to discuss! (@johmedr )

[johmedr]

Thank you for updating this. Seems to work nicely, and can be merged.