Fix the infinite loop in mul_boggle

[LeslieXMOS]

The function mul_boggle will find the closest output scale and bias, and vlashr_shr parameters such that it will not overflow in vpu calculation.

The current code will have an infinite loop when the required vlashr_shr is smaller than vdepth8_shr or (vlsat_shr + vlmul_shr + vdepth8_shr).
As vdepth8_shr and vlsat_shr are unsigned variable, it will underflow when vlashr_shr is too small.
This cause pow(2, y) in L51 and L63 can never output value smaller than 1, thus the loop never ends.

This pull request updated vdepth8_shr and vlsat_shr to signed variable, prevent underflow in the operation.