Do not call assistant in middle of doing multiple tool calls

[attogram]

When a model sends tool_calls with multiple functions

• then first do all tool calls and add results to messages list
• then send the request to the model

AND: watch for model sending index in tool_calls / function, and repeat it back in the tool response.

Currently sends a request after every tool response, slowing things down.
Currently does not send index in tool response.