Without those *.1D files here, I can't test the script. How about if you try it again by removing the line "-censor new_ideal_run1.1d[4] \"? My feeling is that, since you are running 3dDeconvolve with option '-nodata', the censor option could confuse the program, and I guess that is why such an error message pops up.
Regarding the other two questions:
(1) If you switch one of the conditions with your fixation period, that seems to be not a good idea. The reason is, you are fitting the "baseline" with a straight line (-polort 1) or quadratic curve (-polort 2), while the conditions are fitted based on the ideal hrf's generated from waver.
(2) There is probably no common criterion to judge whether a design matrix is good or bad. The condition number of the matrix X'X would give you a general idea about the sensitibiy of the linear system. If the condition number is huge (i.e., 10^9), it is usually a bad design.
If your design matrix is non-invertible, you might consider trying various sets of minlag and maxlag.
Gang