Add distinction between "the two NLPs" in the readme

[matbesancon]

To clarify terminology and as a follow-up on the meeting, we should give terms in the readme between:

• oracle-based NLP with blackbox primal, gradient, hessian
• expression-based NLP

I'm not sure what's the term we agreed upon for the latter

[manuelbb-upb]

From what is in the Google Doc, that would be
• ScalarNonlinearOracleand
• ScalarNonlinearExpression.

I guess, there could also be vector versions of both, so VectorNonlinearOracle and VectorNonlinearExpression?

Concerning the oracles, I don't know if we should assume/require the derivatives to be provided.
This is connected to the task of designing an abstraction layer for the AD system.
Then, if no derivatives are provided for an oracle, we could still do AD if its a pure Julia oracle or maybe finite differences if its truly a blackbox.