This PR addresses #1075 and also fixes a small bug with constant FRD systems with small numbers of frequency points.

Changes:

• Multiplying (*), adding (+), or subtracting (-) a constant from any (MIMO) LTI object now acts element-wise (same as ndarray's). This fixes a bug where multiplying a MIMO LTI system by a constant was multiplying by a matrix filled with the constant rather than a diagonal matrix (scaled identity).
• Specifying an FRD system with fewer than 4 frequency points was generating an error because the default settings try to set up a smooth (interpolating) response and the default degree of the fit was 3. The SciPy interpolation function requires k+1 data points for a degree k fit, so an exception was generated.
• Added unit tests that catch the bug in Multiplying a MIMO TransferFunction by 1.0 multiplies by a matrix of ones, not identity #1075 but also check across all LTI system representations (frd, ss, tf).