Bug: Agent tools lack separate caller-provided runtime context

Problem Agent tool inputs are currently limited by the values available through the model-generated tool-call arguments. This leaves a functional gap for workflows that need to pass caller/runtime context to tools separately from arguments selected by the LLM. For example, a workflow may need a tool to receive customer_id, tenant_id, or session_id from the calling workflow, while the model supplies only a user-intent argument such as order_status. Without a separate mapping, callers must duplicate context in prompts/tool schemas or resolve it through workarounds, and the model may generate values it should not control. Expected behavior • Tool arguments selected by the model are kept separate from values mapped by the calling workflow/runtime. • A calling workflow can map expressions or fixed values into the selected Agent's tool variables. • Caller-provided values take precedence over tool defaults; missing required values produce a clear error. • Mod…