bigint.c: heavy restructuring of big-integers.

Instead of splitting mp_limb by bit operations (using HIGH/LOW macros),
now we use mp_limb2 (which is bigger integer size, e.g uint64_t). It
makes operations (especially mulitiplication) a lot faster. As a side
effect, it also reduces memory consumption (16,452,033 -> 15,545,853 on
my Linux machine).

Other changes:

- trailing zeros are removed after operations.
- division algorithm is simplified.