bellpepper-emulated: Fix overflow calculations

[avras]

The calculation of overflow had some errors. I have added a document in emulated/docs describing the changes. Here are some salient ones.

• EmulatedFieldElement::max_overflow now returns F::CAPACITY as usize - P::bits_per_limb() - 3 instead of F::CAPACITY as usize - P::bits_per_limb()
• The next_overflow for subtraction is changed from a.overflow.max(b.overflow + 2) to a.overflow.max(b.overflow + 1) + 1.
• The padding constant used in sub_padding has changed.
• The limb-wise equality check uses different values for the max_value and number of carry bits.