Using atomic_ref operator= causes compiler error

[Pennycook]

Describe the bug
Although operator= is defined in atomic_ref.hpp, the compiler does not consider it a viable overload.

To Reproduce
Change the test in https://github.com/intel/llvm/blob/sycl/sycl/test/atomic_ref/store.cpp to use operator= instead:

    q.submit([&](handler &cgh) {
      auto st = store_buf.template get_access<access::mode::read_write>(cgh);
      cgh.parallel_for<store_kernel<T>>(range<1>(N), [=](item<1> it) {
        size_t gid = it.get_id(0);
        auto atm = atomic_ref<T, memory_order::relaxed, memory_scope::device,
                              access::address_space::global_space>(st[0]);
        atm = T(gid);
      });
    });

Trying to compile the test gives the following error (repeated several times for each type in the test):

store.cpp:31:13: error: no viable overloaded '='
        atm = T(gid);
        ~~~ ^ ~~~~~~
store.cpp:54:3: note: in instantiation of function template specialization 'store_test<int>' requested here
  store_test<int>(q, N);
  ^
/home/sjpennyc/llvm/build//bin/../include/sycl/CL/sycl/ONEAPI/atomic_ref.hpp:662:7: note: candidate function (the implicit copy assignment operator) not viable: no known conversion from 'int' to 'const sycl::ONEAPI::atomic_ref<int, sycl::ONEAPI::memory_order::relaxed, sycl::ONEAPI::memory_scope::device, sycl::access::address_space::global_space>' for 1st argument
class atomic_ref : public detail::atomic_ref_impl<T, DefaultOrder, DefaultScope,

Environment:

• OS: Linux
• Target device and vendor: All devices
• DPC++ version: Last tested with 66ef4eb
• Dependencies version: N/A

Additional context
We didn't catch this because we don't have tests for all of the operators in atomic_ref. We should add tests for operator T() and operator=(T) at least, since they're intended to be short-hands for load and store (which we do test).

I've tried to fix this myself, but can't make sense of the error. I thought perhaps that operator= needed to be defined in some additional classes in the hierarchy, but nothing I've tried so far has worked. Opening an issue so we don't lose track of this.