Why does adding class specific operator delete activate allocation elision in Visual Studio?

Question

Starting from C++14 we have allocation elision, allowing the implementations to omit calling allocation and deallocation functions. But it was a surprise to me to find what can trigger that kind of optimization in Visual Studio 2022.

In this program

struct A {};
bool x = ( delete new A, 0 );

there is no allocation elision and both new and delete operators are executed in the assembly:

bool x DB     01H DUP (?)

void `dynamic initializer for 'x''(void) PROC
        sub     rsp, 40
        mov     ecx, 1
        call    void * operator new(unsigned __int64)
        mov     edx, 1
        mov     rcx, rax
        call    void operator delete(void *,unsigned __int64)
        mov     BYTE PTR bool x, 0
        add     rsp, 40
        ret     0
void `dynamic initializer for 'x''(void) ENDP

Online demo: https://godbolt.org/z/ozrW9Meqh

But if one defines class-specific operator delete in A:

struct A {
    constexpr void operator delete(void *p) {
        if(p)
          ::operator delete(p);
    }
};
bool x = ( delete new A, 0 );

then allocation elision takes place and assembly code becomes simply

bool x DB     01H DUP (?)

without any dynamic initializer for 'x'. Online demo: https://godbolt.org/z/K4GvnrTT5

Why did adding class specific deallocation function trigger allocation elision? What are other known ways to achieve it?