Updates to Eigen and ptr/reference support with ImplClassTypes and basic types

[maximusron]

• Templated class args like Eigen are resolved to cpp types and Overload matching is succesful in those cases
• The Eigen numba typeinfer is refactored into the C++ expression and handled in numba2cpp
• Returning a reference currently works, have added a passing test for the static reference return
• The pass by reference and pointer support with c_types sets the llvm ir and numba types to ir.PointerType and CPointers respectively in the CppClass data members, beyond typing it faces a sigabort(test11)
• Cling address is now obtained in the CppClassNumbaType via typeof_scope and stored in the RawPointer meminfo, along with the CPPInstance object for cases where the object model is required for lowering
• Eigen test is currently working up until the Boxing call, working on updates using the object model memory info from the native return value when lowering a bitcast instruction

[maximusron]

cc @vgvassilev, @sudo-panda , @wlav

[maximusron]

@wlav

This shouldn't live in the general numba_ext.py. Eigen-specific support is fine, but it should live in its own module that is only imported for folks interested in Eigen. This should be refactored by allowing the registration of hooks (similar to pythonizations or type mapping). As above, "Eigen" stuff should not live in the general module.

I was planning on going ahead by adding a module in cppyy/python with Eigen specific utils that registers the hooks when CppClass instances are detected as Eigen::~ proxies using the __cpp_name__. But instead of requiring users interested in Eigen to import the support module, wouldn't just a conditional import in the general extension work? That way any usage of var = cppyy.gbl.Eigen.~ in the user code, where var is passed in a numba njitted function, would automatically be handled while the general numba extension remains relatively clear of eigen specific code