constant evaluator loses track of most-derived array information when reconstructing `LValue` from `APValue`

[Testcase](https://godbolt.org/z/v8fa8e3Kz):
```c++
struct A { int n; };
struct B : A {} b[2];

constexpr int *f() {
// Loses MostDerived array information.
A *p = b;
return &static_cast(p)[1].n;
}

static_assert(f() == &b[1].n);
```
Clang incorrectly rejects:
```console
:10:15: error: static assertion expression is not an integral constant expression
10 | static_assert(f() == &b[1].n);
| ^~~~~~~~~~~~~~
:7:11: note: cannot access base class of pointer past the end of object
7 | return &static_cast(p)[1].n;
| ^
:10:15: note: in call to 'f()'
10 | static_assert(f() == &b[1].n);
| ^~~
```

This is happening because reconstruction of an `LValue` from an `APValue` (specifically, `SubobjectDesignator` construction) is miscomputing the values for the `MostDerivedIsArrayElement` and `MostDerivedArraySize` fields in the case where the most-derived object that the lvalue refers to is an array element, but the lvalue itself is not (that is, when the lvalue refers to a base subobject of an array element).

That's happening here:

https://github.com/llvm/llvm-project/blob/364bed30971df7fb50d7aadfc21f3996a97bf3a4/clang/lib/AST/ExprConstant.cpp#L238-L242

I think deleting that `else` case would be the proper fix here: for a base subobject, we just want to inherit the "most derived" properties (the "is array" flag and array size) from the previous element.

Reproduce the Godbolt testcase, then inspect clang/lib/AST/ExprConstant.cpp around lines 238-242, focusing on SubobjectDesignator reconstruction and its MostDerived array fields. Verify the fix preserves the array-element information for a base subobject and that the static_assert is accepted.

Written by the indexing model from the issue text.