Fix pandas 3.0 compatibility for groupby aggregation

Change observed=False to observed=True in _group_by_intervals() to fix
test failures on Python 3.11/3.12 with pandas 3.0. The new pandas version
returns all category combinations (including empty ones) when observed=False,
causing aggregation to return hundreds of rows instead of the expected grouped
results.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: michaelchu

optopsy/core.py

@@ -174,7 +174,9 @@ def _group_by_intervals(
 ) -> pd.DataFrame:
     """Group options by intervals and calculate descriptive statistics."""
     # this is a bottleneck, try to optimize
-    grouped_dataset = data.groupby(cols, observed=False)["pct_change"].describe()
+    # Use observed=True to only return groups with actual data (avoids pandas 3.0
+    # issue where observed=False returns all category combinations as empty rows)
+    grouped_dataset = data.groupby(cols, observed=True)["pct_change"].describe()
 
     # if any non-count columns return NaN remove the row
     if drop_na: