This reverts commit 8a68dc7bdd3d1c97677a6633a4f2b5e524c492ae.

This reverts commit d3b6f835d565ec1590059773fc87589ddf8adc37.

This broke the Docker builds and presumably also 32-bit machines that
don't already have libatomic installed.

On some platoforms, 64bit atomic operations need the dedicated helper
library.

This was failing on crossruby, likely because HAVE_GCC_ATOMIC_BUILTINS
was true, but HAVE_GCC_ATOMIC_BUILTINS_64 was false. We probably should
have feature detection of 64-bit stdatomics like we do for GCC, but for
now let's keep rbimpl_atomic_u64_fetch_add in sync with load/set.

Previously this held a pointer to the Fiber itself, which requires
marking it (which was only implemented recently, prior to that it was
buggy). Using a monotonically increasing integer instead allows us to
avoid having a free function and keeps everything simpler.

My main motivations in making this change are that the root fiber lazily
allocates self, which makes the writebarrier implementation challenging
to do correctly, and wanting to avoid sending Mutexes to the remembered
set when locked by a short-lived Fiber.

This only adds the rbimpl_ version to include/ruby/atomic.h so that it
is not a new public interface.

We were already using RUBY_ATOMIC_VALUE_LOAD in a few locations. This
will allow us to use other memory orders internally when desired.

Deduplicates RUBY_ATOMIC_VALUE_LOAD by moving it to ruby_atomic.h.