Ayeaye

Added
- common_pattern.parallel_model_runner.AbstractDependencyDrivenModelRunner to demonstrate how to use
data dependencies to interactively run models.

Updated
- TaskComplete to include the model class' name

Fixed
- Mermaid output as dataset key isn't yet available to annotate the graph edges
- LocalProcessPool which ended up with a deadlock when there are inter-related subtasks. This is beca
use parition_slice is now an iterator, this blocks (surprise!) solution is a thread to manage that as queu
es also need to be managed in parallel.

Fixed
- MultiConnector + wildcards bug reported by burnleyrob - optional_args weren't being passed to the underlying connectors.
- wildcards were not necessarily deterministic. Also reported by burnleyrob

Changed
- methods PartitionedModel.partition_subtask_complete and PartitionedModel.partition_subtask_failed now take a single 'message' instead of multiple arguments describing the task. This will help in the future when traceability is added at the task level.
- improved: doc. around :class:`Model` and :class:`PartitionedModel`
- subclass of `AbstractProcessPool` must accept an iterator instead of a list of sub-tasks. This change makes it possible for a model to interactively send sub-tasks based on the results/completion/failure of other subtasks. Lists of sub-tasks will still work with LocalProcessPool for now. Fossa is being updated to support iterators in parallel.

Fixed
- LocalProcessPool was attempting to set ayeaye.PartitionedModel specifics on ayeaye.Model

Added
- TaskPartition.additional_context so tasks can run with differing resolver contexts to their parent.

Updated
- (kind of a fix) ayeaye.Model.close_datasets only closes datasets that are already open

Updated
- Made it possible for resolver context varibles to result in an empty engine_url. This results
in a `PlaceholderDataConnector`. This can be useful in unittests to switch off datasources.

Fixed
- wildcards in S3 patterns failed to return the bucket name as part of the engine_url

Updated
- warn if readwrite mode is used with S3 files as it isn't cleanly supported (.truncate and .seek) are missing from smart open
- all file based connectors (subs of `FileBasedConnector`) will now auto create directories on writing a file. This was previously just done by the CSV connector is is now consistent.