Code generation for Open Targets data schema

[kpto]

Currently the data schema is hand and hard coded as Python enum. I suggest we introduce code generation for this task which will bring us the following advantages:

1. Easy and quick to support new data version
2. A full data schema of a specific version
3. Prevent human error on data schema
4. Extra code safety on adapter as incorrect schema reference could be caught by code analysis (e.g. referencing outdated fields no longer exist in the new version of data)

An experimental branch will be created. An initial evaluation is being carried out and if it turns out not difficult to implement and everyone is happy with it, this could go into 0.4.0 as well.