`<xstring>`: Fix `basic_string::operator=(const&)` thoroughly

[achabense]

basic_string::operator=(const&) is still problematic even after #3862. For example:

STL/stl/inc/xstring

Lines 3170 to 3175 in 60f1885

	_Traits::copy(_Unfancy(_New_ptr), _Right_ptr, _Right_size + 1);
	_Tidy_deallocate();
	_Mypair._Myval2._Bx._Ptr = _New_ptr;
	_Mypair._Myval2._Mysize = _Right_size;
	_Mypair._Myval2._Myres = _New_capacity;
	} else {

1. After _Tidy_deallocate the string is left in SSO mode, so the pointer should be constructed instead of assigned.
2. The new allocation lacks ASAN annotation.

These are what I can find. As the branch is in such a broken state, I'm uncertain whether there are other problems.

[frederick-vs-ja]

1. After _Tidy_deallocate the string is left in SSO mode, so the pointer should be constructed instead of assigned.

Yes. It seems that there should be test coverage for non-trivial (but constexpr-friendly) fancy pointers.
(I roughly remember that I've invented more than one kinds of fancy pointers for testing.)

[achabense]

there should be test coverage for non-trivial (but constexpr-friendly) fancy pointers.

Fancyptr is implemented and tested in VSO_0000000_fancy_pointers (which does not test special allocators).
Special allocators are implemented and tested in VSO_0000000_allocator_propagation (which does not support fancy ptrs).

Errr I have no idea how to combine these tests gracefully :( Is there a general-purpose helper allocator templace somewhere?