fix(lambda): use FIELDS constant for consistent parquet column order

When creating new monthly parquet files, the hardcoded column list had
wrong order and was missing `pm10`. This caused February 2026 data to
be written with scrambled columns, making the frontend display wrong
metrics (CO2 showing humidity values, etc.).

Now imports `FIELDS` from `storage.py` to ensure consistent column order
across all parquet files.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Author: ryan-williams

src/awair/lmbda/updater.py

@@ -87,10 +87,11 @@ def update_s3_data():
         with atomic_edit(s3_bucket, s3_key, create_ok=True, download=file_exists) as tmp_path:
             tmp_path = Path(tmp_path)
 
-            # If file doesn't exist, create empty DataFrame
+            # If file doesn't exist, create empty DataFrame with correct column order
             if not tmp_path.exists():
                 print('Creating initial parquet file')
-                empty_df = pd.DataFrame(columns=['timestamp', 'temp', 'humid', 'co2', 'voc', 'pm25'])
+                from awair.storage import FIELDS
+                empty_df = pd.DataFrame(columns=FIELDS)
                 empty_df.to_parquet(tmp_path, index=False, row_group_size=DEFAULT_ROW_GROUP_SIZE)
 
             # Use ParquetStorage to manage updates