[FEATURE] WandB query by natural language

[shure-dev]

We expect user query as:

from logllm import wandb_query
wandb_query("get best model when we use chain of thougt")

Then LLM writes
{"order": "-summary.metrics.accuracy", "per_page": 1}
Then execute this

# Query for the best experiment based on accuracy
runs = api.runs(f"{entity}/{project}", {"order": "-summary.metrics.accuracy", "per_page": 1})
best_run = runs[0]

print(f"Best accuracy: {best_run.summary['accuracy']}")
print("Experiment conditions:", best_run.config)

Then we get an answer in very semantic way.