Make _handle_stats_nesting tolerant of missing drop columns

[thodson-usgs]

Summary

_handle_stats_nesting has two .drop(columns=...) calls (one for the geopandas branch, one for the pandas branch) that hardcode literal column names — [\"type\", \"properties.data\"] and [\"data\"] respectively. If a stats response ever comes back in a slightly different shape (renamed key, missing optional key, edge-case feature without the expected nesting), drop() raises KeyError and aborts the helper.

The sibling pd.json_normalize(...) call later in the same function already passes errors=\"ignore\", so this PR adds the same to the two drop() calls for parity.

Diff

# dataretrieval/waterdata/utils.py:894-901
if not geopd:
    df = pd.json_normalize(body[\"features\"]).drop(
        columns=[\"type\", \"properties.data\"], errors=\"ignore\"  # <-- added
    )
    df.columns = df.columns.str.split(\".\").str[-1]
else:
    df = gpd.GeoDataFrame.from_features(body[\"features\"]).drop(
        columns=[\"data\"], errors=\"ignore\"  # <-- added
    )

Test plan

• New test test_handle_stats_nesting_tolerates_missing_drop_columns constructs a stats body whose features lack the top-level type key and confirms the function returns a populated DataFrame.
• Verified that the new test fails on main with KeyError: \"['type'] not found in axis\".
• Full waterdata_utils test suite passes (5 tests).
• Full test suite (excluding deprecated nwis_test.py): 197 passed.

🤖 Generated with Claude Code