`vmull_p64` mismatch with GCC on big-endian

There's a mismatch in the behavior of `vmull_p64` (ARM Neon intrinsic) with GCC on AArch64-BE - clang seems to add an extra `rev64`. [Godbolt](https://godbolt.org/z/j77KW3bcT).
I am not too sure which one's the correct behavior (ARM-BE references are scarce), but I am pretty sure that it shouldn't depend on the compiler lol.

Start with the linked Godbolt reproducer and compare vmull_p64 behavior between Clang and GCC on AArch64 big-endian. Consult the available ARM big-endian references to establish the expected behavior; done means the compiler behavior is consistent and covered by an appropriate regression test.

Written by the indexing model from the issue text.