Add a function call resolution loop and a way to append messages to PromptBuilder

Background

I am working on an automatic ability resolution loop for the WP AI Client in WordPress core (Trac ticket 64865, wordpress-develop PR #12658). It works, but I had to work around two gaps in the PHP AI Client.

@JasonTheAdams suggested in the PR that this logic could live in the PHP AI Client instead:

I wonder if it would be better to introduce agentic looping on function declarations in the PHP AI Client. We could add an event, overloadable function, or some such thing to make it possible for WP to handle Abilities.

The gaps

1. There is no way to append messages to PromptBuilder. withHistory() only prepends. So there is no public way to continue a conversation after a model response. In the WordPress PR, I had to capture the messages and the model from BeforeGenerateResultEvent and call the model directly.

2. There is no built-in loop for function calls. When a model responds with a function call, every consumer has to write the same loop by hand: run the calls, send the results back, and ask the model again. Only the execution of one function call is WordPress-specific. The loop itself is generic and could live in the SDK.

Proposal

A. Add PromptBuilder::withMessages(Message ...$messages) to append full messages, as a counterpart to withHistory().

B. Add a small interface for resolving function calls, and fluent methods to turn the loop on:

$text = AiClient::prompt('What is this site about?')
    ->usingFunctionDeclarations(...$declarations)
    ->usingFunctionCallResolver($resolver)
    ->usingMaxFunctionCallIterations(3) // default 5
    ->generateText();

The loop runs the requested function calls through the resolver, adds the results to the conversation, and asks the model again. It stops when the model answers without function calls, when the resolver cannot handle a call, or after a maximum number of rounds.

I have a working implementation with unit tests and will open a PR with the full details next.

Open questions

1. docs/REQUIREMENTS.md says the client must not include agents. Is a small, bounded loop like this in scope? It builds only on the existing function calling feature.
2. Is an interface the right extension point, or would you prefer an event or an overloadable method?