Cognite-sdk

Added
- The credentials class used in TransformationsAPI, `OidcCredentials`, now also accepts `scopes` as a list of strings
(used to be comma separated string only).

Added
- Missing `unit_external_id` and `unit_quantity` fields on `TimeSeriesProperty`.

Fixed
- Ref to openapi doc in Vision extract docstring
- Parameters to Vision models can be given as Python dict (updated doc accordingly).
- Don't throw exception when trying to save empty list of vision extract predictions as annotations. This is to avoid having to wrap this method in try-except for every invocation of the method.

Added
- Support for new computer vision models in Vision extract service: digital gauge reader, dial gauge reader, level gauge reader and valve state detection.

Added
Support for setting and fetching TimeSeries and Datapoints with "real" units (`unit_external_id`).
- TimeSeries has a new field `unit_external_id`, which can be set when creating or updating it. This ID must refer to a
valid unit in the UnitCatalog, see `client.units.list` for reference.
- If the `unit_external_id` is set for a TimeSeries, then you may retrieve datapoints from that time series in any compatible
units. You do this by specifying the `target_unit` (or `target_unit_system`) in a call to any of the datapoints `retrieve`
methods, `retrieve`, `retrieve_arrays`, `retrieve_dataframe`, or `retrieve_dataframe_in_tz`.

Fixed
- Serialization of `Transformation` or `TransformationList` no longer fails in `json.dumps` due to unhandled composite objects.

Added
- Support for metadata on Workflow executions. Set custom metadata when triggering a workflow (`workflows.executions.trigger()`). The metadata is included in results from `workflows.executions.list()` and `workflows.executions.retrieve_detailed()`.