Hey 👋

I recently came across unasync and wanted to commend you on this incredibly promising & useful project.

In that vein, I wanted to note that certain substitutions (notably additional_replacements={"@pytest.mark.trio": ""}) effectively results in a no-op as a result of the tokenization converting @pytest.mark.trio to the tokens @pytest, mark and trio.
Although token-based execution is theoretically superior, it adds a layer of complexity that httpcore's unasync.py doesn't face.

My recommendation would be to also tokenize the additional_replacements and attempt to match the token streams.
Happy to implement it once we're aligned on a solution.