[Feature] Naming of the domains by an LLM

[quentinblampey]

Currently, domains are annotated as e.g. D1015, and we have to manually provide a name afterwards.

Instead, we could:

1. extract some information from the domains
2. create a prompt
3. send it to a LLM (need an API key)
4. format the output to create a dict[str, str] mapping, e.g. {"D1015": "Tertiary Lymphoid Structure", ...}

Ideas of the information to add to the prompt:

• DEGs between domains
• percentage of cell-types within each domain (if cell_type_key exists).
• tissue name (provided by the user?)
• other ideas?

Function signature:

def llm_domains_name_mapping(
    adata: AnnData,
    obs_key: str,
    cell_type_key: str | None = None,
    tissue_name: str | None = None,
) -> dict[str, str]