Fix demangling of namespaced type arguments

[DutChen18]

The Demangler breaks up a mangled identifier into a list of "scopes":

namespace foo {
	// mangled name: _ZN3foo4funcEv
	// demangled name: foo::func
	// scopes: ["foo", "func"]
	void func() {}
}

This does not work when the idenfitier is for a template with a namespaced type argument:

namespace bar {
	class Arg;
}

namespace foo {
	template<class T>
	void func() {}

	// mangled name: _ZN3foo4funcIN3bar3ArgIiEEEEvv
	// demangled name: foo::func<bar::Arg>
	// scopes: ["foo", "func<bar", "Arg>"]
	template<>
	void func<bar::Arg>() {}
}

The problem is that, when splitting the demangled name into scopes, it ignores the meaning of < and > characters, instead treating them as part of the name. The most hygienic solution is to replace the list of scopes with a more complex structure that can also hold type arguments.

As this would require significant changes to the Demangler, and the Demangler is used in a few places where the effect of such changes would be unclear. I've opted for a quick and dirty solution for now.

The list of scopes will still contain broken names, exactly as shown in the example above. When converting this list of scopes into a javascript name, the templates are removed after joining the scopes together instead of before.

cleanupTemplates used to assume that the template parameters, if any, are at the end of the string passed to cleanupTemplates. This was true before because > is always at the end of a type name, or it is followed by ::. But now, because cleanupTemplates is only called after joining the scopes back together, it can also be passed strings with multiple template parameters, or where the template parameters are not at the end of the string. cleanupTemplates is updated to handle these cases as well, splicing out template parameters from the middle of the string instead of just at the end.

Side note: this whole class is a mess, half of the functions aren't even used anywhere.

[DutChen18]

Some tests seem to be failing, that's what I get for making a PR before running tests.

[DutChen18]

Some tests seem to be failing, that's what I get for making a PR before running tests.

Was a stupid typo, now fixed.

[yuri91]

@DutChen18 add a task to Asana to review this whole class. I agree that it's a mess and when we have time we should rework it properly.
The llvm demangler was bugged at the time, but I think that it works better now. And since we are using more features in client types now it makes sense to make it more robust