Add 32:32->32 division operations (where the lower 32 bits are zero)

[mbitsnbites]

The following two instructions would be useful:

Instruction	Description
DIVHI	Compute (a << 32) / b, signed
DIVHIU	Compute (a << 32) / b, unsigned

If the first source operand is larger than the second source operand, an overflow will occur. This should produce the same result as division by zero (as specified for DIV and DIVU).

One use case for these instructions is to implement fixed point division, such that:

    div     s3, s1, s2    ; s3 = 32-bit integer part of division (a / b)
    rem     s4, s1, s2    ; s4 = division remainder (a % b)
    divhi   s4, s4, s2    ; s4 = 32-bit fractional part of division (((a % b) << 32) / b)