Rework NLP docs section: classical NLP vs search/retrieval models

### Problem

The [NLP](https://www.elastic.co/docs/explore-analyze/machine-learning/nlp) section (and its children) no longer matches what it contains.

It originally covered classical NLP / text analytics (NER, text classification, zero-shot, extractive QA, fill-mask, language identification, plus HF/Eland deploy workflows). Over time it also absorbed search/retrieval models such as ELSER, E5, Elastic Rerank, and the Jina family (embeddings, multimodal, rerankers, reader)—which are not classical NLP tasks.

That naming and IA makes it harder to find true NLP content, and overlaps with Elastic Inference and Solutions → Search docs for the same retrieval models.

Related: #7603

### Possible directions (high level)

Either keep a renamed/broader models section under Machine learning, **or** split classical NLP from search/retrieval models (for example toward Elastic Inference / Search). Exact scope and IA TBD.