Limit bit-rot in the disabled backends

[dra27]

While getting the other-checks stage to pass again in #11183, I disabled the CI checks for the presently disabled native code backends. While they may not be working, I think it's probably worth having make check_all_arches at least limit the bit-rot by ensuring that they continue to be compile, even if the code they produce is nonsense.

This PR:

• Refactors uses of Iload for the changes in Introduce atomic loads in Cmm and Mach IRs #10943 in all the disabled backends
• Removes a rebase artefact that accidentally reverted part of More warnings when compiling the compiler. #502 in the arm backend
• Fixes the use of Iextcall in the arm emitter for (type) compatibility with multicore
• Applies the renaming of Domainstate.Domain_exception_pointer to Domainstate.Domain_exn_handler between Move C global variables to a dedicated structure #8713 and the multicore merge to the i386 backend
• Adds assert false placeholders for Idls_get to all the disabled backends

Together, these changes restore make check_all_arches

[dra27]

I don't think we've made a final decision on the fate of the two 32-bit backends, which is why I've included them in the CI list for check_all_arches but it would also be reasonable to consider just adding power, s390x and riscv (i.e. the 64-bit architectures which we do intend restoring support for)

[nojb]

LGTM: ensuring compilability of all backends is very good, and the changes won't cause hidden bugs whenever work on them resumes.