Deeplake

- Fixed dataset size when concurrently deleting the same row.

Improvements in version-control, new data types and observability
- Improved experience with version control and tagging
- Allow merging branches
- Integrated with opentelemetry for better observability
- Added `float16` and `bfloat16` data types

Big release bringing major features to deeplake users:
- [Allow creating branches in deeplake datasets](https://docs.deeplake.ai/latest/api/misc/#branching)
- [Allow running queries with directly passing credentials to the storage](https://docs.deeplake.ai/latest/api/query/#deeplake.query)
- Support all JOIN types and cross product in deeplake query engine
- Improve labelbox integration. Allow authenticating by IAM roles

Additionally, small fixes and improvements are done:
- Optimize the way deeplake stores [dataset metadata](https://docs.deeplake.ai/latest/api/misc/#dataset-metadata)
- Allow MAXSIM comparison on dynamic columns

Fixed bugs:

- Permanent redirect error message from S3 client on different location resource loading
- Incorrect dtype in virtual columns
- Stuck or crash when doing multiple iterations with break
- Descriptive error message on network request failures

- Fix for high memory usage on json column queries
- Improvement in json data and keys querying
- Allow appending array of nones to image type columns

- Improved deeplake memory usage during queries
- Improved dataset load time and memory usage
- Max Marginal Relevance search native support with TQL `MMR_SIMILARITY` function
- New index type EmbeddingMatrix, to improve `MAXSIM` search
- Added default value to `add_column`
- Support for integrating with labelbox
- Series of bugfixes and quality improvements