Clean up internal LLM API (#484)

* Fresh diff

* remove instructionhandler

* updated internal interface to make all tests pass

* fixed caching tests

* updated llm.ipynb

* removed unnecessarily defensive validation

* updated tool call decoding to use concrete type of tool result instead of annotations

* updated completions to fix basic type errors

* updated call assistant to handle decoding tool calls

* dropped stale comments

* moved model and param model back to internals of `completions`

---------

Co-authored-by: Kiran Gopinathan <kiran@basis.ai>

Author: eb8680

docs/source/agent.py

@@ -3,8 +3,9 @@
 from effectful.handlers.llm import Template
 from effectful.handlers.llm.completions import (
     LiteLLMProvider,
-    compute_response,
-    format_model_input,
+    Message,
+    call_assistant,
+    call_user,
 )
 from effectful.ops.semantics import fwd, handler
 from effectful.ops.syntax import defop
@@ -31,28 +32,27 @@ def wrapper(self, *args, **kwargs):
                 with handler(
                     {
                         Agent.current_agent: lambda: self,
-                        format_model_input: self._format_model_input,
-                        compute_response: self._compute_response,
+                        call_user: self._format_model_input,
+                        call_assistant: self._compute_response,
                     }
                 ):
                     return template(self, *args, **kwargs)
 
             setattr(cls, method_name, wrapper)
 
-    def _format_model_input(self, template, other, *args, **kwargs):
+    def _format_model_input(self, template, env):
         # update prompt with previous list of messages
         prompt = fwd()
         if Agent.current_agent() is self:
-            assert self is other
             self.state.extend(prompt)
             prompt = self.state
         return prompt
 
     def _compute_response(self, *args, **kwargs):
         # save response into persisted state
-        response = fwd()
+        response: Message = fwd()
         if Agent.current_agent() is self:
-            self.state.append(response.choices[0].message.model_dump())
+            self.state.append(response)
         return response