Replicate input structure in output files

[angus-g]

The output file is simply a concatenation of the filtered, flattened arrays for each timestep:

HDF5 "test.h5" {
GROUP "/" {
   DATASET "var_U" {
      DATATYPE  H5T_IEEE_F32LE
      DATASPACE  SIMPLE { ( 2, 70000 ) / ( 2, 70000 ) }
   }
}
}

It would be good to copy the structure of the input file (spatial and temporal dimensions) to the output file.

[angus-g]

The most straightforward approach would probably be to open the xarray DataArray for each sampled variable, and combine them into a single Dataset. However, this interface doesn't support incremental writes. If we can set the Dataset's data to a lazy dask array, we could get around this?

[angus-g]

pydata/xarray#1672 pertains to the feature in xarray that would make this easier.

[angus-g]

Closed via #14