
1)
The bsim noise C code was broken, going back to at least 4.7.0.  The
correlated noise slot was not initialized to zero in tnoimod != 2. 
See spice3f5-wr for an example fix.

The noise reference results here, and in 4.7.0, are incorrect.

After fixing, WRspice and spice3f5 agree very closely.

2)
The S/D swap test shows a large difference in dcSweep_nrd and dcSweep_nrs.
This is OK, the test is bogus as we should be swapping nrd/nrs values
as well as s/d nodes.
