Fix depol error param for device noise model with depol + relaxation error

[chriseclectic]

Summary

Fixes #703

Details and comments

Running example code in issue 703 now returns:

Gate Infidelities (relax + depol, depol, relax)
id[0] = [0.00076, 0.00042, 0.00076]
u2[0] = [0.00076, 0.00042, 0.00076]
u3[0] = [0.00152, 0.00085, 0.00152]
id[1] = [0.00041, 0.00041, 0.00014]
u2[1] = [0.00041, 0.00041, 0.00014]
u3[1] = [0.00083, 0.00083, 0.00029]
id[2] = [0.00045, 0.00045, 0.00014]
u2[2] = [0.00045, 0.00045, 0.00014]
u3[2] = [0.0009, 0.0009, 0.00028]
id[3] = [0.00057, 0.00057, 0.0002]
u2[3] = [0.00057, 0.00057, 0.0002]
u3[3] = [0.00115, 0.00115, 0.0004]
id[4] = [0.00061, 0.00061, 0.00036]
u2[4] = [0.00061, 0.00061, 0.00036]
u3[4] = [0.00122, 0.00122, 0.00071]
cx[0, 1] = [0.01053, 0.00789, 0.01053]
cx[1, 0] = [0.01159, 0.00789, 0.01159]
cx[1, 2] = [0.00754, 0.00754, 0.00218]
cx[1, 3] = [0.01122, 0.01122, 0.0058]
cx[2, 1] = [0.00754, 0.00754, 0.00252]
cx[3, 1] = [0.01122, 0.01122, 0.00539]
cx[3, 4] = [0.00755, 0.00755, 0.00509]
cx[4, 3] = [0.00755, 0.00755, 0.00575]

So the relax + depol error is approx equal to the max error of the depol only or relax only error case.