Refactor to reuse stream parsing across ChatModels (#380)

* Copy discard_none_arguments to LitellmChatModel

* Switch OpenaiChatModel to use Stream approach

* Add TODO to test retry logic

* Add StreamParser to DRY openai streaming code

* Add _if_given

* Fix typing for openai Stream classes

* Add parse_stream and use in OpenaiChatModel

* tidy function schema matching

* Copy Stream classes into stream.py

* Switch LitellmChatModel to use stream parsers

* Switch OpenaiChatModel to use shared streaming classes

* Add docstring for is_instance_origin

* Use type origins in parse_stream

* Fix complete type hints for LitellmChatModel

* Change function_schemas type list -> Iterable

* Add StreamState and OpenaiStreamState

* Add TODOs

* Update openai retry test cassettes

* Add back usage for OpenaiChatModel

* Consolodate parsers into one

* Make LitellmChatModel use new parsing format

* Fix litellm_ollama

* Handle multiple tools in a chunk, for Mistral

* Remove anthropic context manager usage

* Add _if_given helper for anthropic

* Allow parser.get_content to return None

* Switch AnthropicChatModel to new parsing logic

* Delete unused validate_str_content functions

* Remove redundant TODOs

* Fix prompt_chain and unskip tests

* Only yield tool call args if not falsy

* Add FunctionCallNotAllowedError, ObjectNotAllowedError

* Add UnknownToolError, raise in OutputStream

* Remove done todo for unknown tool call

* Remove is_content_ended if favor of is_tool_call

* Add typecheck to make all

* Add get_function_schemas. Fix some mypy errors

* Tidy calculation of allow_string_output

* Fix remaining mypy errors

* Delete is_instance_origin

Makefile

@@ -56,4 +56,4 @@ docs-serve:  # Build and serve the documentation
 	uv run mkdocs serve
 
 .PHONY: all
-all: format lint test
+all: format lint typecheck test