refactor: ditch Nominatim, use reverse_geocoder exclusively for geocoding

Nominatim was slow (rate-limited, 1 req/sec, ~4s timeouts) and unreliable
from Docker/Cloudflare (blocked or throttled). It was the root cause of
location labels failing to appear during trip scrubbing.

- server/app.py: _geocode_cached now uses reverse_geocoder exclusively
  (offline, instant, city/province level). Removed Nominatim HTTP calls,
  ThreadPoolExecutor, concurrent.futures import.
- app.py (Pi): same — reverse_geocoder imported at module level,
  Nominatim + urllib.request removed.
- Batch endpoint simplified to sequential loop with lru_cache + dedup.
  No network dependency at all — geocoding is instant and always works.

reverse_geocoder provides city/province labels (e.g. 'Montreal, Quebec')
which is the right granularity for trip scrubbing.

server/app.py

@@ -2,7 +2,6 @@
 uploaded by the Pi, storing them in Postgres + on-disk file storage.
 Also serves the remote web portal for browsing and managing the configured car.
 """
-import concurrent.futures
 import functools
 import hashlib
 import os
@@ -21,45 +20,28 @@ from modules.user_auth import login_manager, User, check_password, require_login
 from modules import device_config, device_registry
 
 # ── Geocode cache ──────────────────────────────────────────────────────────
-try:
-    import reverse_geocoder as _rg
-    _RG_AVAILABLE = True
-except ImportError:
-    _RG_AVAILABLE = False
+# reverse_geocoder is an offline Python library that maps lat/lon →
+# city/province using a precomputed gazetteer (~40MB). Instant, no network,
+# always available — a much better fit for trip scrubbing than Nominatim
+# which is slow, rate-limited, and often unreachable from Docker/Cloudflare.
+import reverse_geocoder as _rg
 
 
 @functools.lru_cache(maxsize=512)
 def _geocode_cached(lat4, lon4):
-    """Try Nominatim first (street-level), fall back to reverse_geocoder (city),
-    then raw coordinates. Always returns a non-empty label string — never None."""
+    """Reverse-geocode a coordinate to city/province using the offline
+    reverse_geocoder library. Always returns a non-empty label string."""
     lat = lat4 / 10000.0
     lon = lon4 / 10000.0
     try:
-        import urllib.request, urllib.error, json as _json
-        url = (
-            f"https://nominatim.openstreetmap.org/reverse"
-            f"?lat={lat}&lon={lon}&format=jsonv2&accept-language=en"
-        )
-        req = urllib.request.Request(url, headers={"User-Agent": "yaris-obd-server/1.0"})
-        with urllib.request.urlopen(req, timeout=4) as resp:
-            data = _json.loads(resp.read())
-        a = data.get("address", {})
-        road = a.get("road") or a.get("hamlet") or a.get("suburb") or ""
-        city = a.get("city") or a.get("town") or a.get("village") or a.get("municipality") or ""
-        label = ", ".join(p for p in [road, city] if p)
-        if label:
-            return label
-    except Exception:
-        pass
-    # Offline fallback
-    if _RG_AVAILABLE:
         results = _rg.search([(lat, lon)], mode=1, verbose=False)
         if results:
             r = results[0]
             label = ", ".join(p for p in [r.get("name", ""), r.get("admin1", "")] if p)
             if label:
                 return label
-    # Last resort — raw coordinates so the frontend never gets an empty label
+    except Exception:
+        pass
     return f"{lat:.5f}, {lon:.5f}"
 
 
@@ -629,16 +611,14 @@ def geocode():
 def geocode_batch():
     """Batch geocode — returns {key: label} for all unique coords in one request.
 
-    Parallelised with ThreadPoolExecutor (max_workers=4) so Nominatim HTTP
-    calls and reverse_geocoder lookups run concurrently instead of sequentially.
-    _geocode_cached's lru_cache absorbs duplicate coords within the batch."""
+    Uses the offline reverse_geocoder library (instant, no network). Deduplicates
+    coords before lookup; lru_cache absorbs any remaining duplicates."""
     body = request.get_json(silent=True) or {}
     coords = body.get("coords")
     if not isinstance(coords, list) or not coords:
         return jsonify(error="expected {coords: [{lat, lon}, ...]}"), 400
 
-    # Deduplicate — same (lat4,lon4) would hit lru_cache anyway, but this
-    # avoids submitting redundant futures.
+    # Deduplicate
     seen = set()
     items = []
     for item in coords:
@@ -654,19 +634,8 @@ def geocode_batch():
             items.append((f"{lat:.3f},{lon:.3f}", lat4, lon4))
 
     result = {}
-    if not items:
-        return jsonify(result)
-
-    with concurrent.futures.ThreadPoolExecutor(max_workers=4) as ex:
-        futures = {ex.submit(_geocode_cached, lat4, lon4): key for key, lat4, lon4 in items}
-        for future in concurrent.futures.as_completed(futures):
-            key = futures[future]
-            try:
-                result[key] = future.result(timeout=8)
-            except Exception:
-                # Last resort — raw coords if even the fallback failed
-                parts = key.split(",")
-                result[key] = f"{float(parts[0]):.5f}, {float(parts[1]):.5f}"
+    for key, lat4, lon4 in items:
+        result[key] = _geocode_cached(lat4, lon4)
 
     return jsonify(result)