ARM: fix ffs/fls implementations to match x86

ARMs ffs/fls implementations are not type compatible with x86, so when
they're used in combination with min()/max(), they provoke warnings.
Change these to be inline functions with the correct types, providing
the clz as a separate documentation, and document their individual
behaviours.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>