use fast, approximate functions for overlap/correction algo speed increase

The majority of the time of the calo reco is now spent on shower overlap and correction calculations.

This MR introduces very fast approximations of exp, log, sqrt for corrections speedup from LHCb::Math::Approx. Will increase speed of overlap/correction algorithms with 20-30%.

In addition, all these calculations should be done with floats, not doubles as is now, as the precision needed is smaller than floating-point precision. Note: in the CondDB the parameters for these corrections are stored as doubles, but will be converted during initialization in the correction algos now.