MiddlewareCallContext

What a middleware receives. Contains:

• params: the in-going call params, exactly as the upstream invoker constructed them. Middleware may modify and pass a new copy to doGenerate / doStream.

• model: the wrapped LanguageModel (i.e., this middleware's enclosing wrapper). Exposed for metadata reads such as model.modelId. Do NOT call model.generate(p) / model.stream(p) from inside a middleware — that re-enters the chain from the top and infinite-loops. Use doGenerate / doStream instead, which targets the rest of the chain past this middleware.

• doGenerate: invoke the downstream chain (the next middleware's wrapGenerate, or the underlying model's generate if this is the innermost middleware).

• doStream: invoke the downstream chain on the streaming axis.

Both doGenerate and doStream reference the SAME downstream depth — so a middleware's wrapStream calling context.doGenerate(...) skips its own wrapGenerate and invokes the rest of the chain's generate path. This is the load-bearing property that lets simulateStreamingMiddleware work.