Fix ARM implementation of mulusmall_loop_up(digit, ptr, len, newdigit)...
...by removing the shortcut for small 'digit' and calling
mulu32_64_vregs instead (which handles digits >= 2^16 fine).
The requirement that 'digit' be small (<= 36) is documented in cl_DS.h
but nowhere enforced. Indeed, it is clearly violated in the base-N
input function digits_to_I_baseN().
I am not sure what to do. This ARM asm seems to be the only code which
relies on small 'digit'. A future patch should either fix the input
method to guarntee that 'digit' is small or accept that it can be
large and remove the shortcut in mulusmall_loop_up for good (and also
fix the comments in cl_DS.h).