M2-T04: add single-row record CRUD API (patch/delete)

- PATCH/DELETE /api/locations/{person}/{datetime} and /api/poo/{timestamp}
- update only non-PK fields (PK immutable); 404 on missing PK
- delete scoped to exact full PK with rowcount guard (0->404, 1->ok);
  no batch/truncate/drop path
- session + CSRF protected; bare ingestion endpoints untouched
- service helpers in app/services/location.py and poo.py; regenerate openapi/
- tests/test_api_record_crud.py

app/services/poo.py

@@ -4,7 +4,7 @@ from dataclasses import dataclass
 from datetime import datetime, timezone
 import logging
 
-from sqlalchemy import desc, insert, select
+from sqlalchemy import delete, desc, insert, select
 from sqlalchemy.orm import Session
 
 from app.config import Settings
@@ -74,6 +74,53 @@ def record_poo(
             logger.warning("Failed to trigger poo webhook on Home Assistant: %s", exc)
 
 
+def update_poo_record(
+    session: Session,
+    timestamp_pk: str,
+    *,
+    status: str | None,
+    latitude: float | None,
+    longitude: float | None,
+) -> PooRecord | None:
+    """Update non-PK fields of a single poo record row.
+
+    Returns the updated ORM object, or ``None`` if the PK does not exist.
+    The ``timestamp`` PK is immutable and must not be passed as an update field.
+    Only fields with a non-``None`` value are written.
+    """
+    row = session.execute(
+        select(PooRecord).where(PooRecord.timestamp == timestamp_pk)
+    ).scalar_one_or_none()
+
+    if row is None:
+        return None
+
+    if status is not None:
+        row.status = status
+    if latitude is not None:
+        row.latitude = latitude
+    if longitude is not None:
+        row.longitude = longitude
+
+    session.commit()
+    session.refresh(row)
+    return row
+
+
+def delete_poo_record(session: Session, timestamp_pk: str) -> bool:
+    """Delete the single poo record row identified by its PK.
+
+    Returns ``True`` if exactly one row was deleted, ``False`` if the PK did
+    not exist (caller should raise 404).  The DELETE is scoped to the exact PK
+    — no batch/truncate path exists.
+    """
+    result = session.execute(
+        delete(PooRecord).where(PooRecord.timestamp == timestamp_pk)
+    )
+    session.commit()
+    return result.rowcount == 1
+
+
 def get_latest_poo_record(session: Session) -> LatestPooRecord | None:
     stmt = select(PooRecord).order_by(desc(PooRecord.timestamp)).limit(1)
     record = session.execute(stmt).scalar_one_or_none()