fix: address second round of Copilot review comments

- Fix date→datetime promotion: use isinstance checks instead of
  hasattr(tzinfo) which doesn't work for plain date objects
- Use param.default_unit for unit fallback instead of hardcoded ug/L
- Parse BDD AnalysisDate as date (not datetime) to match DB column type
- Guard after_scenario cleanup to skip DB session when no fixtures created

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: kbighorse

tests/features/environment.py

@@ -768,10 +768,10 @@ def after_scenario(context, scenario):
 
     # Chemistry backfill cleanup — runs regardless of DROP_AND_REBUILD_DB
     # because the backfill steps create their own fixture data.
-    if getattr(context, "_backfill_created", None) is not None:
+    created = getattr(context, "_backfill_created", None)
+    if created is not None and any(created.values()):
         try:
             with session_ctx() as session:
-                created = context._backfill_created
 
                 # Delete in FK order: Notes → Observations → Samples → FieldActivities → FieldEvents → NMA rows
                 # Scope to observations linked to this scenario's samples

tests/features/steps/chemistry-backfill.py

@@ -822,12 +822,12 @@ def _create_minor_trace_from_fields(context: Context, fields: dict):
             session.flush()
             context._backfill_created["nma_sampleinfo_ids"].append(sampleinfo.id)
 
-        # Parse analysis date
+        # Parse analysis date as a plain date to match the Date column type
         analysis_date = None
         if fields.get("AnalysisDate"):
             analysis_date = datetime.strptime(
                 fields["AnalysisDate"], "%Y-%m-%d"
-            ).replace(tzinfo=timezone.utc)
+            ).date()
 
         # Parse numeric fields
         sample_value = (

transfers/backfill/chemistry_backfill.py

@@ -398,30 +398,30 @@ def _backfill_minor_trace_chemistry_impl(session: Session) -> BackfillResult:
         # Build observation_datetime — observation_datetime is NOT NULL in the DB,
         # so rows without AnalysisDate must be skipped.
         obs_dt = row.analysis_date
-        if obs_dt is not None:
-            if hasattr(obs_dt, "tzinfo") and obs_dt.tzinfo is None:
-                from datetime import datetime, time
-
-                if isinstance(obs_dt, datetime):
-                    obs_dt = obs_dt.replace(tzinfo=timezone.utc)
-                else:
-                    # date object — promote to datetime at midnight UTC
-                    obs_dt = datetime.combine(obs_dt, time.min, tzinfo=timezone.utc)
-        else:
+        if obs_dt is None:
             result.errors.append(
                 f"Row GlobalID={global_id_str} has no AnalysisDate — skipping"
             )
             continue
 
-        # unit is NOT NULL in the DB — fall back to parameter default_unit
-        unit = row.units if row.units else "ug/L"
+        from datetime import date as date_type, datetime, time
+
+        if isinstance(obs_dt, datetime):
+            if obs_dt.tzinfo is None:
+                obs_dt = obs_dt.replace(tzinfo=timezone.utc)
+        elif isinstance(obs_dt, date_type):
+            obs_dt = datetime.combine(obs_dt, time.min, tzinfo=timezone.utc)
+        # else: already timezone-aware datetime — use as-is
 
         savepoint = session.begin_nested()
         try:
             param = _get_or_create_parameter(
                 session, analyte, "water", default_unit="ug/L"
             )
 
+            # unit is NOT NULL in the DB — fall back to parameter's default_unit
+            unit = row.units if row.units else param.default_unit
+
             analysis_method_id = None
             if row.analysis_method:
                 am = _get_or_create_analysis_method(session, row.analysis_method)