[929][evalution/io] Make lead time and source windew available to evaluation via OutputDataset (#1087)

* store lead time in OutptDataset and make it available to evaluation

* ruffed

* addressed comments

* remove lead_time_hrs from OutputDataset

* use type alias for union of `zarr.Array` and `NDArray`

* dataclass to store time range information

* add source interval to `OutputDataset`

* store lead-time and source interval in xarray

* correct de(serialization) for `OutputDataset` with source-interval

* correctly instantiate `OutputDataset` from `OutputBatchData`

* remove attribute `t_window_len_hours` from `OutputBatchData`

* calculate for output source windows from sample indices

* add source_intervals as attribute to OutputBatchData

* ruffed

* fix: pass source intervals to OutputBatchData

* fix: use correct string in array.astype

* fix: (de)serialize np.datetime64

* fix deserialization of OutputDataset from json/zarr

* fix: errors in xarray conversion

* fix types

* Improve documentation

* remove lead_time from OututDataset

* implement with_target function for ItemKey class

* handle potentially missing target/prediction data in OutputItem

* separate extraction of targets/predictions into method

* include fstep 0 when forecast offset is 1

* Infer forecast offset in ZarrIO

* pass forecast offset to OutputItem

* use forecast_offset in OutputItem

* exclude fsteps without targets from iteration

* ruffed

* make forecast offset property instead of attribute

* raise meaningfull error if ZarrIO is empty

* fix: pass key instead of zarr.group

* ruffed

Author: grassesi