Cognite-sdk

Added
- `complete` parameter on `datapoints.retrieve_dataframe`, used for forward-filling/interpolating intervals with missing data.
- `include_aggregate_name` option on `datapoints.retrieve_dataframe` and `DatapointsList.to_pandas`, used for removing the `|<aggregate-name>` postfix on dataframe column headers.
- datapoints.retrieve_dataframe_dict function, which returns {aggregate:dataframe} without adding aggregate names to columns
- source_created_time and source_modified_time support for files

Added
- New method create_hierarchy() added to assets API.
- SequencesAPI.list now accepts an asset_ids parameter for searching by asset
- SequencesDataAPI.insert now accepts a SequenceData object for easier copying
- DatapointsAPI.insert now accepts a Datapoints object for easier copying
- helper method `cognite.client.testing.mock_cognite_client()` for mocking CogniteClient
- parent_id and parent_external_id to AssetUpdate class.

Changed
- assets.create() no longer validates asset hierarchy and sorts assets before posting. This functionality has been moved to assets.create_hierarchy().
- AssetList.files() and AssetList.events() now deduplicate results while fetching related resources, significantly reducing memory load.

Added
- files.create() method to enable creating a file without uploading content.
- `recursive` parameter to raw.databases.delete() for recursively deleting tables.

Changed
- Renamed .iteritems() on SequenceData to .items()
- raw.insert() now chunks raw rows into batches of 10,000 instead of 1,000

Fixed
- Sequences queries are now retried if safe
- .update() in all APIs now accept a subclass of CogniteResourceList as input
- Sequences datapoint retrieval updated to use the new cursor feature in the API
- Json serializiation in `__str__()` of base data classes. Now handles Decimal and Number objects.
- Now possible to create asset hierarchy using parent external id when the parent is not part of the batch being inserted.
- `name` parameter of files.upload_bytes is now required, so as not to raise an exception in the underlying API.

Added
- Variety of useful helper functions for Sequence and SequenceData objects, including .column_ids and .column_external_ids properties, iterators and slice operators.
- Sequences insert_dataframe function.
- Sequences delete_range function.
- Support for external id column headers in datapoints.insert_dataframe()

Changed
- Sequences data retrieval now returns a SequenceData object.
- Sequences insert takes its parameters row data first, and no longer requires columns to be passed.
- Sequences insert now accepts tuples and raw-style data input.
- Sequences create now clears invalid fields such as 'id' in columns specification, so sequences can more easily re-use existing specifications.
- Sequence data function now require column_ids or column_external_ids to be explicitly set, rather than both being passed through a single columns field

Fixed
- Renamed Model.schedule_data_spec to Model.data_spec so the field from the API will be included on the object.
- Handling edge case in Sequences pagination when last datapoint retrieved is at requested end
- Fixing data points retrieval when count aggregates are missing
- Displays unexpected fields on error response from API when raising CogniteAPIError

Added
- Support for model hosting exposed through experimental client

Fixed
- Handling dynamic limits in Sequences API