otbcli_DimensionalityReduction: Add options to output eigenvalues, apply inverse transform for PCA and correctely deal with no-data

Add options to otbcli_DimensionalityReduction

1. Output eigenvalues along eigenvectors for PCA
2. Apply the inverse transform (I understand this is already done using the "-outinv" parameter.
3. Deal with nodata
4. Option to estimate stats from a certain % sampled from the image
5. Option to output eigenvalues and eigenvectors only, avoiding calcalating and writing the components (useful to decide how many components to keep in a second run that would perform the actual calculation of the components)