Add a chunk in the NCReader

[kongdd]

Suggest adding a chunk in the struct of NCReader.
In my running case (10 years hourly runoff simulation), 90% time consumed at loading forcing data.

Adding a chunk (with the dimension of nlon*nlat*chunksize, chunksize=1000) to store forcing data should reduce the nc reading time significantly.

If the running date is still within the current chunk, read data from the chunk; otherwise, update the chunk and read again.

Wflow.jl/src/io.jl

Lines 623 to 630 in 6a28144

	struct NCReader{T}
	dataset::CFDataset
	dataset_times::Vector{T}
	cyclic_dataset::Union{NCDataset,Nothing}
	cyclic_times::Dict{Tuple{Symbol,Vararg{Symbol}},Vector{Tuple{Int,Int}}}
	forcing_parameters::Dict{Tuple{Symbol,Vararg{Symbol}},NamedTuple}
	cyclic_parameters::Dict{Tuple{Symbol,Vararg{Symbol}},NamedTuple}
	end

[verseve]

Thanks for the suggestion. We generally run with forcing data that has a chunksize of 1 (time dimension), then ~2% of total simulation time is used for loading forcing (and cyclic) data (daily time step).

Do you see a reduction in time consumption when a chunksize of 1 is used in the forcing data?

If not, it would be good to know the Wflow.jl version (and preferably the NCDatasets version) used, Operating System and ideally a test case (part of your running case).

[kongdd]

Thanks for your solution. The running times is reduced about 10 times.

setting chunksize to 1 (time dimension) might need to be documented, which will benefit other users.