lib: count_zeros: unify count_{leading,trailing}_zeros()

lib: count_zeros: unify count_{leading,trailing}_zeros()

The 'leading' helper returns BITS_PER_LONG if x == 0, while 'trailing'
one returns COUNT_TRAILING_ZEROS_0, which turns to be -1.

None of the current users explicitly check the returned value for
COUNT_TRAILING_ZEROS_0, except the loongarch, which tests implicitly
for the '>= 0'.

So, align count_trailing_zeros() with the count_leading_zeros(), and
simplify the loongarch handling.

Reviewed-by: Andy Shevchenko <andriy.shevchenko@intel.com>
Signed-off-by: Yury Norov <ynorov@nvidia.com>