## Breaking
- Rename task `cookfarm` to `cookfarm_mlr3`.
This was done to distinguish the `cookfarm` task implementation in {mlr3} better from the original `cookfarm` dataset.
`cookfarm_mlr3` also now comes with all rows of the upstream `cookfarm` task and not with a random subset as before.
- Rewrite `mlr_resampling_spctcv_cstf` implementation.
**The method will produce different fold results compared to v1**.
This is because of a change/fix in the sampling behavior: before, an (unwanted) stratified sampling was done on `time` and `space` variables.
While this matched the upstream implementation in {CAST}, this did not match with the actual theoretical underpinning of the method found in the literature.
## Features
- Add support `DataBackendVector` backend.
- Add support for `DataBackendRaster` (@be-marc, #191).
- `mlr_resampling_spctcv_cstf`: a log message returns the column roles from the Task which are used for partitioning
- The help pages for all methods now describe the methods manually rather than importing the upstream documentation of the respective method.
## Bugfixes
- All {mlr3spatiotempcv} methods now comply with the {mlr3} man file declaration logic.
## Misc
- Escape all examples and tests for non-installed packages.
- The `cookfarm_mlr3` task now sets column roles "space" and "time" for variables `SOURCEID` and `Date`, respectively.
- Harden CLUTO tests (#182)