Add application-level extension field and PrepareCall hook to AgentCall / AgentStreamCall

[fwang2002]

Currently AgentCall and AgentStreamCall only carry a fixed set of fields (sampling params, messages, tools, callbacks, etc.). There is no first-class way for application-layer code to:

1. Attach arbitrary, application-specific data to a call.
2. Run a one-shot preparation hook before the first model invocation, analogous to the internal agent.prepareCall.

Typical use cases for the hook: translating an application-defined context object into SystemPrompt / ProviderOptions / Messages, injecting tenant-scoped headers, resolving feature flags, hydrating RAG context, etc.

[fwang2002]

Additionally, the createPrompt flow should also be taken into account. In particular, the system prompt (and potentially other fields in agent settings) needs to be dynamically generated or overridden based on the result returned by the PrepareCall / preparation hook, rather than being limited to static configuration at call creation time.

A typical scenario is: when all prompts are managed by Langfuse, the agent settings only need to specify the prompt path (or prompt ID). The PrepareCall hook can then automatically fetch the actual prompt content at runtime.