feat: Phase 2 — leadership 大屏 (/overview) + district normalization + drawer a11y

Phase 2 of the UX modernization. Three conflict-free workstreams.

Leadership 驾驶舱 (/overview):
- Wuhan 13-district Leaflet choropleth (public/wuhan_districts.geojson, keyed
  on name, darker=higher per 高风险高亮), legend, hover/click-zoom
- 全部/门诊/住院 Segmented toggle drives choropleth + Top-5 district bar
- literal "数据截至2023-12" as-of badge (D3 honesty); raw spinner → LoadingState
- decompose OverviewDashboard 501→273; 6 components + 2 helpers under components/overview/

District normalization (backend data boundary):
- case_loader.normalize_district + load_cases_by_district_daily collapse the
  26 dirty labels (武昌/武昌区…) → 13 canonical; analysis/grid/insights repointed
  (fixes a grid-merge row-drop bug as a bonus); in-memory, schema unchanged

Shell a11y (code-review carryover):
- drawer is now a proper modal: ESC, body scroll-lock, focus-in + focus-trap
  cycle + focus-restore, role=dialog/aria-modal/aria-label, hamburger aria-expanded
- SideNav expanded state lifted to AppShell so rail+drawer stay in sync
- RouteErrorBoundary around <Outlet/> keeps shell chrome on page/chunk failure

Gates: tsc 0 · vitest 64 · e2e 19/19 (17 user-flows + 2 overview) · build ok
· backend pytest 6 new + 48 regression green

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

backend/data/case_loader.py

@@ -25,12 +25,42 @@ _cache: dict[str, Optional[pd.DataFrame | datetime]] = {
     "outpatient": None,
     "inpatient": None,
     "combined": None,
+    "cases_by_district_daily": None,
     "loaded_at": None,
 }
 
 # Guards the lazy build so concurrent callers don't duplicate the load/concat.
 _load_lock = threading.RLock()
 
+# Canonical Wuhan administrative districts (13), matching the `name` field in
+# Datas/武汉市.geojson. All district roll-ups must collapse to exactly these.
+CANONICAL_DISTRICTS = [
+    '江岸区', '江汉区', '硚口区', '汉阳区', '武昌区', '青山区', '洪山区',
+    '东西湖区', '汉南区', '蔡甸区', '江夏区', '黄陂区', '新洲区',
+]
+# Bare (suffix-less) base name -> canonical 区-suffixed name.
+_DISTRICT_BASE_TO_CANONICAL = {d[:-1]: d for d in CANONICAL_DISTRICTS}
+_DISTRICT_SUFFIXES = ('区', '县', '市')
+
+
+def normalize_district(name: str) -> str:
+    """Map a district label to its canonical 区-suffixed form.
+
+    The case parquet carries both bare ("武昌") and suffixed ("武昌区") spellings
+    of the same district, which double-counts in any roll-up. This collapses
+    them: known bare names map to their canonical form; already-suffixed names
+    pass through unchanged; anything else gets a "区" appended.
+    """
+    if name is None:
+        return name
+    name = str(name).strip()
+    if name in _DISTRICT_BASE_TO_CANONICAL:
+        return _DISTRICT_BASE_TO_CANONICAL[name]
+    if name.endswith(_DISTRICT_SUFFIXES):
+        return name
+    return f"{name}区"
+
+
 # Wuhan district mapping
 WUHAN_DISTRICTS = {
     '江岸区': ['江岸'],
@@ -170,3 +200,40 @@ def get_inpatient_data() -> pd.DataFrame:
     """Return the cached inpatient dataframe"""
     load_data()
     return _cache["inpatient"]  # type: ignore[return-value]
+
+
+def load_cases_by_district_daily() -> pd.DataFrame:
+    """Load processed/cases_by_district_daily.parquet with districts normalized.
+
+    The on-disk parquet carries both bare and 区-suffixed spellings of each
+    district (26 labels = 13 districts × 2 spellings), so any groupby on the
+    raw `district` column double-counts. This is the single data-access
+    boundary: it normalizes labels to the canonical 13 and re-aggregates
+    (sum of outpatient_count / inpatient_count / total_cases per
+    normalized district + date), so every downstream consumer
+    (analysis / grid / insights) sees clean, deduped 13-district data.
+
+    Returns a copy with columns [date, district, outpatient_count,
+    inpatient_count, total_cases]. Raises FileNotFoundError if the parquet
+    is missing (callers handle this as they did before).
+    """
+    path = PROCESSED_DIR / "cases_by_district_daily.parquet"
+    cached = _cache.get("cases_by_district_daily")
+    if cached is not None:
+        return cast(pd.DataFrame, cached).copy()
+
+    with _load_lock:
+        cached = _cache.get("cases_by_district_daily")
+        if cached is not None:
+            return cast(pd.DataFrame, cached).copy()
+
+        df = pd.read_parquet(path)
+        df["district"] = df["district"].map(normalize_district)
+        agg = (
+            df.groupby(["date", "district"], as_index=False)[
+                ["outpatient_count", "inpatient_count", "total_cases"]
+            ]
+            .sum()
+        )
+        _cache["cases_by_district_daily"] = agg  # type: ignore[assignment]
+        return agg.copy()