fix(dashboard): keep numeric dtype so columns sort numerically (#46)

The summary table coerced metrics to formatted strings via f'{x:.2f}',
which made the Gradio DataFrame sort them lexicographically
(114.42, 1785.64, 181.58 …). Use pd.to_numeric(...).round(2) instead
so the underlying dtype stays numeric and sorting behaves as expected.

Fixes #24

Author: Hugoch

extra/dashboard/app.py

@@ -77,7 +77,7 @@ def summary_table() -> pd.DataFrame:
              'token_throughput_secs']]
         for metric in ['inter_token_latency_ms_p90', 'time_to_first_token_ms_p90', 'e2e_latency_ms_p90',
                        'token_throughput_secs']:
-            data[metric] = data[metric].apply(lambda x: f"{x:.2f}")
+            data[metric] = pd.to_numeric(data[metric], errors='coerce').round(2)
         data = data.rename(
             columns=column_mappings)
         return data