fix: non-type template (enum) instantiation for templ func

[Vipul-Cariappa]

requires test:

import cppyy
from cppyy import gbl

cppyy.cppdef(
    r"""
enum What { NO, YES };

template <What E>
struct EE {
    What w = E;
};

template <What E>
What get() {
    return E;
}
"""
)

assert gbl.EE[gbl.What.NO]().w == 0
assert gbl.EE[gbl.What.YES]().w == 1

assert gbl.get[gbl.What.NO]() == 0
assert gbl.get[gbl.What.YES]() == 1

This was discovered while working on ROOT workshop slides.

[Vipul-Cariappa]

Added the test to compiler-research/cppyy#164

[vgvassilev]

Why do we need to do string parsing to do a simple-ish lookup.

[Vipul-Cariappa]

This is required for now. There is already some string parsing through the Cpp::GetScopeFromCompleteName and Cppyy::GetScope, I wanted to just reuse them instead of introducing this new function. But that is not possible. All of the *GetScope* fail to resolve enum constants. If I fix that, then other tests fail, because that is the expected behaviour according to cppyy.

We should eventually be able to reduce string parsing, but not eliminate it entirely.
To reduce the string parsing, we will need to update https://github.com/compiler-research/CPyCppyy/blob/9d95a1954d0285f85087e40d638b48b40558b42d/src/TemplateProxy.cxx#L175-L176 not to construct a string, instead return std::vector<Cpp::TCppType_t>, or something similar. I believe all the necessary type information is already stored as QualType*, we unnecessarily convert QualType* to std::string, and back to QualType* again.
I will open an issue to track it.
But we cannot remove the string parsing completely because users can instantiate templates with string. Example: func["int", "double", "MyClass"](1, 2.0, gbl.MyClass(...))

[Vipul-Cariappa]

Issue opened at compiler-research/CPyCppyy#146

[vgvassilev]

Sounds good. Let's move forward!

[vgvassilev]

LGTM!