VpDivd bugfix

[tompng]

Fix VpDivd bug which is a blocker of #371.

Fix rounding: VpDivd should always truncate. Remove wrong optimization that uses VpAsgn which rounds with the current rounding mode.

VpDivd(div, rem, a, b)
a=0.9999999999 9999999999 9999999999 999000000E3 (1, 5, 5)
b=0.1E1 (1, 1, 1)

# Before
div=0.1000E4 (1, 1, 3) # Wrong. It shouldn't be rounded with the current rounding mode.
rem=0.0 # Wrong. It shouldn't be zero.

# After
div=0.9999999999 99000000E3 (1, 3, 3)
rem=0.9999999999 99000000E-18 (-2, 2, 6)
rem=0.9999999999 9999999999 9000000E-9

Fix VpDivd loop bug that sometimes skip calculating last DECDIG.

# OK case
VpDivd(div, rem, a, b)
a=0.3000000000 0000000000 0000000000 0000000000 0000000000 0000000000 0001999999 999E73 (9, 9, 10)
b=0.10E2 (1, 1, 1)

div=0.3000000000 0000000000 0000000000 0000000000 0000000000 0000000000 0001999999 99E72 (8, 8, 10)
rem=0.9E1 (1, 1, 10)

# Bug case
VpDivd(div, rem, a, b)
a=0.3000000000 0000000000 0000000000 0000000000 0000000000 0000000000 0000999999 999E73 (9, 9, 10)
b=0.10E2 (1, 1, 1)

# Before. a, b, div, rem has same precision as OK case, but the last DECDIG calculation is somehow skipped
div=0.300000000E72 (8, 1, 10)
rem=0.999999999E9 (1, 1, 10)

# After
div=0.3000000000 0000000000 0000000000 0000000000 0000000000 0000000000 0000999999 99E72 (8, 8, 10)
rem=0.9E1 (1, 1, 10)

[tompng]

I want to add a test for VpDivd, but the current VpDivd calculation is weird.

VpDivd(c, r, a, b);
c->frac[0] // This value is always zero.

I'll add a test in a separate pull request with VpDivd fix.