Publish documentation

[kaaveland]

We should have something bare-bones on rtd, explaining the basic philosophy and design:

• Minimal amount of code that works
• Speed advantage by catering directly to pyarrow dataset API
• "Bring your own DatalakeServiceClient", if you can create it, we can use it -- which means we support all authentication that azure-storage-filedatalake supports

Then we could have some examples for common things like reading hive partitioned datasets, "arrow advantages" like self_destruct, strings_to_categorical etc.