C++: Use Definition.qll's getName

[jbj]

It turns out we didn't have to move the getName implementation into the mirror classes in QualifiedName. Doing so only made it harder for the optimiser to specialize calls to getName on various kinds of Declaration.

Most of the diff in this PR (the first commit) is a partial revert of #1303, reverting just the parts that cause unnecessary complication. The cost is that there are now references in both directions between QualifiedName.qll and Declaration.qll.

Specialization of getName still doesn't work as well as I would like, but this change does show some improvement. For example, current master has this DIL:

EscapesTree::stdIdentityFunction#f(pure unique Function::Function f) :-
  exists(pure Namespace::Namespace call_result |
    exists(pure Namespace::Namespace receiver |
      exists(pure Namespace::Namespace call_result#4 |
        exists(pure string arg1 | arg1 = "std", namespaces(call_result#4, arg1)),
        Declaration::Declaration::getNamespace#bf(f, call_result#4)
      ),
      (
        exists(string arg1 |
          arg1 = "move", Declaration::Declaration::getName_dispred#bb(f, arg1)
        );
        exists(string arg1 |
          arg1 = "forward",
          Declaration::Declaration::getName_dispred#bb(f, arg1)
        )
      ),
      Declaration::Declaration::getNamespace#bf(f, receiver),
      Namespace::Namespace::getParentNamespace_dispred#ff(receiver, call_result)
    ),
    exists(pure string arg1 | arg1 = "", namespaces(call_result, arg1))
  )
.

while this PR gets us this DIL:

EscapesTree::stdIdentityFunction#f(pure unique Function::Function f) :-
  exists(pure Namespace::Namespace call_result |
    exists(pure Namespace::Namespace receiver |
      exists(pure Namespace::Namespace call_result#4 |
        exists(pure string arg1 | arg1 = "std", namespaces(call_result#4, arg1)),
        Declaration::Declaration::getNamespace#bf(f, call_result#4)
      ),
      (
        exists(pure string arg1, pure dontcare int _ |
          arg1 = "move", functions(f, arg1, _)
        );
        exists(pure string arg1, pure dontcare int _ |
          arg1 = "forward", functions(f, arg1, _)
        )
      ),
      Declaration::Declaration::getNamespace#bf(f, receiver),
      Namespace::Namespace::getParentNamespace_dispred#ff(receiver, call_result)
    ),
    exists(pure string arg1 | arg1 = "", namespaces(call_result, arg1))
  )
.

The only difference is the disjunction that matches "move" and "forward".

[geoffw0]

I've done a few performance checks and what I tried seems to be as quick, or slightly quicker with these changes. Difficult to tell what's going on in the 'Clause timing report' though.

[geoffw0]

This was difficult to review because there's a fair bit of history explaining these changes (@jbj filled me in with some of the details on Slack). I won't pretend to be 100% happy but (1) most of this PR is reverting recent changes that were deemed to be a mistake (2) I've tested and it seems to behave well in practice and (3) there is good test coverage to ensure correctness.