Bug: source_prompt loaded incorrectly in evaluation script (string indexed as list)

[marghovo]

In the evaluation script, the source prompt is loaded with:

src_prompt = data_dict["source_prompt"][0].replace("[", "").replace("]", "")

However, source_prompt is a string, not a list. Indexing with [0] on a string
returns only the first character (e.g., "A" instead of the full prompt).

Fix:

src_prompt = data_dict["source_prompt"].replace("[", "").replace("]", "")