Modify BaseProblem.data_misfit to include data covariance matrix

[jwhhh]

This happens when data_misfit is set to be L2 and calculated by us via residual.

1. If the data covariance matrix isn't set, use identity matrix by default
2. If the data covariance matrix is diagonal, use the following formula to improve performance $A^T A$ where 
   $A = \sqrt{C_d^{-1}}\times{res}$ where $\times$ means element by element
3. If the data covariance matrix isn't diagonal, use ${res}^TC_d^{-1}{res}$ instead