Mindsdb

Fixes
- Hugging Face required parameters (e.g. 'task') are not checked before model is created
- Mongo:
- create predictor from mongo: integration and query are optional
- join model and table in mongo: query is optional
- Fix imports for new ML module dataprep_ml
Changes
- check model metadata before creation
- pytorch has to be in tags
- metadata.pipeline_tag equal model task
- Improved base error message if model fails post-download

Fix for HF:
- data_integration_ref can be None

- fix for 'retrain' of old models

New features:
- Managing model versions: set active, querying particular version and other
- Tagging of models

New data handlers:
- Vitess
- OpenGauss

Other:
- improved error messaging in some cases
- improved reading of csv files
- various bigfixes and improvements

- fix for 'RETRAIN MODEL'

- Migrating to `projects` structure.
`Project` - is virtual database containing models and views. `mindsdb` database becomes a `project`. Please see documentation for more information about `projects`.
- new version of Lightwood as default ML engine
- сhanging working with database, logs, configs to be able run components independently
- massive documentation update.
- various improvements and bugfixes.

**New data handlers:**
- Databend
- Solr
- Impala
- Google Sheets
- TDEngine
- Orioledb
- StarRocks
- Oceanbase

**New ML engines:**
- Huggingface
- Merlion