make an explicit cast to field type for shape-function values to allow float128 ctype grids. Note: the problem is that the runtime-order lagrange finite-elements do not support different range-type than double internally.