Gordo

* Don't convert DataFrame to ndarray in cv (732)
* Bump gunicorn from 20.0.3 to 20.0.4
* Bump pytest from 5.3.0 to 5.3.1
* Bump pip-tools from 4.2.0 to 4.3.0

* Convert workflow component unittests to pytest (697)
* Add default_asset to sensortag and dataset (699)
* Add NoSuitableDataProviderError (702)
* Refactor model building with a ModelBuilder (689)
* Add new asset to path, 1904, 1170 and 1219 (701) NOTE: no path known datalake specified, so loading will fail.
* Watchman list then watch, model update interval 20m -> 60m (708)
* Add thread lock to data_provider get_client (707)
* Restarting k8s watch on error events (710)
* Support anomaly factors in DiffBasedAnomalyDetector (673)

- Add SensorTagNormalizationError with exit code 60 (684)
- Workflow: Clean up quoting data to builder (683)
- Use Model CRD for model build notification (680) (note: this means you need a cluster with an updated version of infrastructure installed)
- Workflow: Perform manual pod GC at workflow exit (695)

Model building:
- Add ability to scale y when scoring models (599)
- Support setting random seed in evaluation key (632)
- Accept metrics list in evaluation config key (631)
- Model builder: Add different exit codes for different exceptions (668)
- Patch model's dataset config with global dataset config (656)

Client:
- Support configuring client Session via CLI (622)
- Fix client's metadata handling (621)
- Client.predict(..) optionally take specific endpoints (626)

Server:
- Add access and error logfile to server (625)

Workflow generator:
- Add checks for invalid model config in workflow generator (630)

K8s scalability fixes:
- Set model builder requests=limits, cpu: 1001, memory:3900
- Add exponential retries to workflow resource steps (638)
- Workflow: delete pods when workflow succeeds (645)

Changes

* Rename Dockerfile-Argo -> Dockerfile-GordoDeploy (614)
_Dockerfile-Argo isn't reflective to what it is/does_
* Remove custom keras model serializations (609)
_No longer create a tree representing a pickled pipeline,
where any Keras models are saved with .h5 file. Rather
make our Keras models pickle-able and pickled the entire
model/pipeline directly._
* Mark test_argo_link with dockertest
* Ignore benchmarks in 'test' setup.py command
* Add cpu limits to misc utility workflow steps
* Add limits to postgres and grafana
* Remove bad chars in path
_Remove invisible chars in the path as they are removed from the datalake._
* Remove unused tqdm from requirement files
* pip-tools upgrade test_requirements.txt
* pip-tools upgrade requirements.txt
* Consolidate CI tests into ~4-5min run times (602)
_Consolidate the CircleCI tests into ~4-5min run times,
which introduces no code changes, only CIrcleCI config_
* Remove httpbin fixture (600)
_Not use by tests any longer_

Requirement updates

No changes from [0.38.0-rc1](https://github.com/equinor/gordo-components/releases/tag/0.38.0-rc1)