refscale: Add non-atomic per-CPU increment readers

refscale: Add non-atomic per-CPU increment readers

This commit adds refscale readers based on READ_ONCE() and WRITE_ONCE()
that are unprotected (can lose counts, "refscale.scale_type=incpercpu"),
preempt-disabled ("refscale.scale_type=incpercpupreempt"),
bh-disabled ("refscale.scale_type=incpercpubh"), and irq-disabled
("refscale.scale_type=incpercpuirqsave").  On my x86 laptop, these are
about 4.3ns, 3.8ns, and 7.3ns per pair, respectively.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>