Documentation improvements

[podcastinator]

Starting this thread to track and collect various documentation improvements.

• Document initialization APIs for client and enclave, and explain the various client / enclave keys
• Add config file description for init_client API; add it to the demos as well
• Consensus v. non-consensus mode, and which APIs require consensus
• Clarify that obliviousness only supported for hist training at present
• Text input format; e.g. https://xgboost.readthedocs.io/en/latest/tutorials/input_format.html