[conv.lval] Add example of erroneous 'trap representation' being read

[Eisenwave]

Fixes #7047.

[Eisenwave]

Now presumably based on CWG2899.

If CWG2899 is accepted, this example would need to be shifted up by one bullet, and it may make sense to extend the example to show how this happens without erroneous values, like in the issue.

[jensmaurer]

I take this as a request to add an example to CWG2899. Done.

[Eisenwave]

@jensmaurer thanks, it makes sense to add it as an example there.

However, the reason why I've originally suggested the example was that it isn't obvious how erroneous values can trigger this form of UB, so ideally, CWG2899 should show both scenarios (erroneous and non-erroneous cause):

bool f() {
+  bool e;            // e has erroneous value (6.7.4 [[basic.indet]](https://wg21.link/basic.indet))
+  bool u = e;        // undefined behavior if the value of e is not a valid value representation for bool
+
  bool b = true;
  char c = 42;
  memcpy(&b, &c, 1);  // implicitly creates objects at b (23.5.3 [[cstring.syn](https://wg21.link/cstring.syn)], 6.7.2 [[intro.object](https://wg21.link/intro.object)])
  return b;           // undefined behavior if 42 is not a valid value representation for bool
}