[3.1?] AI Gateway: Document semantic association of MCP servers based on prompts

## Description

Add documentation describing how to semantically select MCP servers based on the incoming prompt, instead of exposing all tools to the model by default. Include details about:

* Dynamically exposing only relevant MCP servers per prompt
* Restricting capabilities (read-only when appropriate)
* Limiting tools based on OIDC roles

## Definition of Done (TBD)

* New section explaining semantic MCP server selection behavior (possibly in the AI MCP Proxy plugin)
* Plugin Examples showing:
* Per-prompt server selection
* Restricting write tools when intent does not justify them
* OIDC role-based tool exposure
* How-to - this is a nice-to-have. If you don't manage to do it, create a ticket for it, and assign it to me.

## Information

Person of contact: @spacewander
Implementation details: TBA

## Size

S to M