feat(matchers): match by phone number and email address

Cargo.toml

@@ -31,6 +31,7 @@ once_cell = "1.21.4"
 osm_pbf_iter = { version = ">=0.2.3", default-features = false }
 parquet = { version = "60.0.0", default-features = false, features = ["arrow", "geospatial", "zstd"] }
 parquet-geospatial = { version = "60.0.0", default-features = false }
+phonenumber = "0.3.10"
 piz = "0.5.1"
 prost = "0.14.4"
 protobuf_iter = { version = "0.1.3", default-features = false }
@@ -45,6 +46,7 @@ serde_json = { version = "1.0.150", default-features = false, features = ["std"]
 tempfile = "3.27.0"
 time = { version = "0.3.53", default-features = false, features = ["formatting", "parsing", "std"] }
 tokio = { version = "1.52.3", default-features = false }
+unicode-normalization = "0.1.26"
 uuid = { version = "1.24.0", default-features = false, features = ["v4", "v5"] }
 webpki-roots = "1.0.6"
 wkb = "0.9.2"

deny.toml

@@ -28,6 +28,13 @@ ignore = [
   # release picks it up, then drop these two lines.
   "RUSTSEC-2026-0194", # quadratic run time checking duplicate attribute names
   "RUSTSEC-2026-0195", # unbounded namespace-declaration allocation (OOM)
+  # atomic-polyfill is unmaintained, not vulnerable. It comes in via
+  # phonenumber -> postcard -> heapless 0.7, which only uses it on targets
+  # without native atomics, so it is never built for the x86_64 and aarch64
+  # targets we ship (`cargo tree -i atomic-polyfill` prints nothing there;
+  # only `--target all` shows it). Drop this line once postcard moves to
+  # heapless 0.8, or phonenumber stops using postcard.
+  "RUSTSEC-2023-0089",
 ]
 
 # ── licenses: only accept licenses we've reviewed as OK to redistribute ─────

docs/TECHNICAL_DESIGN.md

@@ -346,7 +346,9 @@ fetch away.
   in spatial order; for each feature, builds a search cap around its
   centroid and queries `OsmFeatureIndex` for OSM candidates inside it,
   scoring each (see [`src/matchers/`](../src/matchers/)) and keeping
-  the best. Writes one row per ATP feature — matched or not — to
+  the best. Opens the nondescript phone and email sets from
+  `import_atp`, so that hotlines and central mailboxes never count as
+  matching evidence. Writes one row per ATP feature — matched or not — to
   `conflated.parquet`. See
   [`docs/outputs/CONFLATED_PARQUET.md`](outputs/CONFLATED_PARQUET.md)
   for that file’s schema. **~8 minutes** (~5 min matching, ~3 min
@@ -603,10 +605,13 @@ until the full pipeline was proven out:
 - Nothing yet uploads suggested edits to any of the tools described
   above — that integration (most likely MapRoulette first, given its
   API is the closest fit) is future work, not yet designed in detail.
-- The only matcher implemented, `PoiMatcher`, matches shop-category
-  POIs solely on an exact `brand:wikidata` tag. We’d like to extend
-  this well beyond it: matching stores by name too, not just
-  `brand:wikidata`, and matching well beyond stores (a tree matcher
+- The only matcher implemented, `PoiMatcher`, matches shops,
+  restaurants, hotels, schools and fuel stations on an exact
+  `brand:wikidata` tag, or on a shared phone number or email address
+  with a compatible category (see
+  [`notes/2026-10-25-phone-and-email-matching.md`](notes/2026-10-25-phone-and-email-matching.md)).
+  We’d like to extend this well beyond it: matching by name too, and
+  matching well beyond points of interest (a tree matcher
   conflating municipal tree datasets against OSM by spatial distance
   and species looks particularly tractable). See
   [#708](https://github.com/brawer/osmdiffs/issues/708).