Add support for reduction using dimension values in split-apply-combine pattern

[ph-kev]

In the split-apply-combine, reductions are supported by Reduction. It would be useful to also pass information about the dimension values to the reduction. The syntax might looks like

time_averaged_var =
    var |>
    ClimaAnalysis.GroupAll("time") |>
    ClimaAnalysis.ReduceWithDims(integrate) |>
    ClimaAnalysis.combine

and the signature of the reduction could be function integrate(data, dim_name, dims_values; dims) end where data is being reduced over, dim_name is the name of the dimension, dims_values are all the dimension values (in the form of a dictionary mapping dimension name to dimension values), and dims is the index to do the reduction on.

The use case for this is to support operations like finding the argmax/argmin or integrating time series data.

One concern I have is whether it makes sense to pass in more information like the attributes and dimension attributes. Although, there are work around for this by defining a functor for the reduction that store the attributes and dimension attributes in the OutputVars.

[ph-kev]

It might be good to pass it as a struct or named tuple, so that more fields can be added without breaking the interface.