`filter_nt` reallocation

[juliasloan25]

Is your feature request related to a problem? Please describe. https://github.com/CliMA/ClimaLSM.jl/pull/431 fixed a bug in the state FieldVector structure, but also introduced a function filter_nt. This function takes in a NamedTuple, and returns a new NamedTuple (reallocating) with all key/value pairs with value nothing are removed. This is needed because initialize_vars now maps a model component with no prognostic/auxiliary variables to nothing. When a CanopyModel has components that aren't prognostic (e.g. with prognostic hydraulics but not conductance), we need to filter out these components before constructing the FieldVector for Y, to avoid nothings in Y and maintain compatibility with ClimaTimesteppers.

filter_nt is used in the CanopyModel methods of initialize_prognostic and initialize_auxiliary, and duplicates the NamedTuples produced by initialize_vars in these functions. This is not ideal because all of these values are reallocated. We should come up to an alternative solution to CanopyModel components not having prognostic variables that avoids this reallocation.