v1.2.0: Add C++ implementation of standard error calculations

 - Add C++ implementation of standard error calculations. This implementation
   uses SVD rather than a full pseudo-inverse.
 - Add argument `se.smooth` for experimental Laplace smoothing on the response
   probabilities when doing standard error calculations. This provides better
   numerical stability and avoids very small standard errors.
 - The original poLCA code is used for `nclass=1`
 - Bump to C++20
 - No OpenMP to stop overthreading in the regression model
 - Minor update to the documentation
 - Minor improvements in the C++ code by using more of the arma and C++
   features such as using in-place operations and containers instead of raw
   arrays