v0.7.2

Fixed

• FuelWatch (au_fuelwatch) ignored the search radius — the WA FuelWatch RSS
  endpoint has no native radius parameter, so the provider returned every
  station in the WA Region regardless of the user's radius_km setting.
  async_list_stations now applies a great-circle (haversine) filter
  client-side when lat, lng and radius_km are supplied, so the station
  picker shows only stations within the configured radius. Reported in #44.
• e-control (at_econtrol) ignored the search radius — same class of bug:
  the API hard-caps to 10 nearest stations and exposes no radius parameter,
  so radius_km was silently a no-op. async_list_stations now applies a
  client-side haversine filter, sharing haversine_km from the new
  providers/_geo module with au_fuelwatch.
• Empty station list silently created a broken entry — for location-search
  providers (e.g. au_fuelwatch), choosing coordinates outside the
  provider's coverage area (Sydney + 1 km on a WA-only feed, or a radius
  too small to capture any station) caused the config flow to fall through
  to entry creation with no station_id. The runtime then synthesised a
  station_id from lat/lng (e.g. au_fuelwatch_-33.86880_151.20930) that no
  provider could resolve, leaving the entry stuck in
  Failed setup, will retry: Station '…' not found in FuelWatch feed.
  The config flow now loops back to the previous step (location or
  county) with a no_stations_found_location / no_stations_found
  error banner so the user can widen the radius or pick a different
  county without restarting the flow from the country picker. The free-text
  station_id fallback that allowed any string through is gone. National-
  average / global_list providers (which genuinely have one synthetic
  entry) keep the silent-create shortcut; the EU Oil Bulletin path is
  unchanged.
• National-average providers asked for coordinates — Albania, Czech
  Republic, Malta, Moldova, Montenegro, Netherlands and Poland (ORLEN)
  publish only a single national reference row; the config flow
  nonetheless prompted for lat/lng/radius on setup and then discarded
  them. These providers now use STATION_LOOKUP_MODE = "global_list"
  (matching EU Oil Bulletin), so the location step is skipped and the
  user goes straight from provider → station picker → entry creation.

Changed

• Re-enabled six providers verified live against their real upstreams
  during this audit: al_fuel (Albania), cz_ccs (Czech Republic),
  md_fuel (Moldova), mt_fuel (Malta), nl_anwb (Netherlands) and
  pl_benzyna (Poland / ORLEN). README status flipped from ⚠️ Disabled
  to 🤖 Smoke-tested for each. The remaining six providers in the
  "upstream broken" bucket (ba_fuel, dk_fuelfinder, es_minetur,
  fi_tankille, lu_carbu, pt_dgeg) still fail their live probe —
  these stay disabled.

Internal

• New providers/_geo.py module with a shared haversine_km function plus
  a filter_within_radius helper. Extended with an optional get_coords
  callable so providers with nested coord shapes (at_econtrol,
  au_nsw, au_qld, au_vic, ch_tcs) use the same helper as
  flat-coord providers. All 16 providers that previously carried a
  private _haversine_km copy (au_nsw/au_qld/au_vic/be_carbu/ca_qc/
  ch_tcs/es_minetur/fr_carburants/gb_fuelfinder/ie_pumps/is_fuel/it_mase/
  pt_dgeg/se_bensinpriser/si_goriva/no_drivstoff) now import from
  providers._geo. Providers whose filter loop matched the helper's
  drop-on-missing-coords contract collapsed their loop into a single
  filter_within_radius(...) call (at_econtrol, au_fuelwatch, au_nsw,
  au_qld, au_vic, ch_tcs, ie_pumps, pt_dgeg). The rest (be_carbu, ca_qc,
  es_minetur, fr_carburants, gb_fuelfinder, is_fuel, it_mase,
  no_drivstoff, se_bensinpriser, si_goriva) kept their inline loop —
  each has provider-specific gating (e.g. keep-on-missing-coords,
  late-extracted station IDs) that differs from the helper's strict
  contract — and only swapped the haversine math to _geo.haversine_km.
  The duplicate base.haversine_km (atan2 formula, numerically
  identical to within ~1e-12 km) is deleted. No behaviour change.
• Dropped unused latitude/longitude/radius_km constructor parameters
  from national-average / no-coords providers (al_fuel, ba_fuel,
  dk_fuelfinder, eu_oil_bulletin, lt_saurida, md_fuel, me_fuel,
  mt_fuel, pl_benzyna) — these providers return a single country-level
  row (or have no per-station GPS at the source) and never read coordinates.
  Also dropped the unused county constructor parameter from ba_fuel,
  dk_fuelfinder and mt_fuel, and the **kwargs absorber from
  me_fuel.__init__ so kwarg typos now surface as TypeError.
• Aligned the radius_km=0 contract across every client-side filter
  provider: 0 / None / missing kwarg = "no filter" (matches
  providers._geo.filter_within_radius's falsy-check semantics).
  Twelve providers used kwargs.get("radius_km") or self._radius_km,
  which silently rewrote an explicit 0 back to the constructor default
  (au_nsw, au_qld, au_vic, be_carbu, ca_qc, ch_tcs, de_tankerkoenig,
  es_minetur, fr_carburants, pt_dgeg, se_bensinpriser, si_goriva); three
  more (at_econtrol, au_fuelwatch, ie_pumps) used
  kwargs.get("radius_km", default) which preserves 0 but differs
  cosmetically. All fifteen now use the same
  kwargs["radius_km"] if kwargs.get("radius_km") is not None else <default>
  ternary that preserves a user-supplied 0. ca_qc.__init__ and
  pt_dgeg.__init__ also dropped their radius_km or 10.0 init
  rewrites in favour of the strict is not None check.
• Added the no_stations_found / no_stations_found_location /
  no_stations_found_global keys to strings.json and every locale's
  config.abort block (matching the existing config.error entries) so
  HA picks up the translated text on the re-rendered location / county
  step and on the global_list abort path.