[AI Gateway] Clarify Guardrails streaming evaluation behavior (#32964)

* [AI Gateway] Clarify Guardrails streaming evaluation behavior

* Address review feedback: model-type caveat, active voice, remove duplicated streaming note

* Distinguish REST API vs legacy gateway streaming guardrail behavior

* Lead streaming section with unsupported framing

* Address style-guide feedback: prose over two-item list, standard link phrasing

* Use neutral 'gateway endpoints' phrasing per docs terminology

* Update src/content/docs/ai-gateway/features/guardrails/usage-considerations.mdx

Co-authored-by: Ming Lu <mlu@cloudflare.com>

* [AI Gateway] Clarify Guardrails response evaluation

---------

Co-authored-by: Ming Lu <mlu@cloudflare.com>