test: unit tests for helpers (openai_compatible_helpers, async_helpers)

[planetf1]

Add unit tests for the helpers module — 77% overall but with critical gaps in code used by every OpenAI-compatible backend.

Scope

File	Coverage	What's missing	Est. tests
openai_compatible_helpers.py	~12%	extract_model_tool_requests(), chat_completion_delta_merge(), message_to_openai_message() — all pure functions, zero tests	~15
async_helpers.py	~28%	send_to_queue(), wait_for_all_mots() — needs async fixtures/mocks	~6
server_type.py	~46%	_server_type() — pure URL parsing, trivial	~3

event_loop_helper.py is at ~52% and the untested paths are complex error-handling branches — lower priority.

Notes

• openai_compatible_helpers.py is the headline — three pure functions on the critical path for all OpenAI/Ollama/LiteLLM backends, completely untested at unit level
• chat_completion_delta_merge() merges streaming chunks — good candidate for canned data testing
• Part of test: improve test coverage across formatters and core #813

~24 tests, single coherent module.