Support asynchronous execution of `Template.__call__`

LLM API invocations are slow and many applications have significant parallelism between invocations that we should be able to exploit. 

If we choose to use Python's async support, we should avoid leaking async behavior into the rest of `effectful`. This means that the core functions in `effectful.ops` should not become async functions and we should not add async duplicates of these functions.