Draft: Use ERFA for coordinate transformation instead of casacore
Doing a preliminary analysis of the timings of the predict step in DP3. Perf suggested that this part of the code is the one in which is spent more then 60% of the execution. This is probably due to the many indirection present in casacore measures. As suggested by Tammo Jan we could convert to use ERFA in this part of the code. As it delivers very similar precision but it does not contains mutex (as in casacore).
In this merge request the switch to using erfa is incorporated with the addition of some additional useful routines in case further transformation have to be performed.