Bug 1958992 - suggest: Improve geonames l10n and weather-suggestions matching.

This is a substantial reworking of geonames and weather suggestions in suggest.

Summary of major changes:

In RS, don't store geonames' alternate names inline with the core geonames data.
Instead, use separate record types. (As a reminder, "alternates" just means
variants of a geoname's main name, like "NYC" and "NY" are alternates for New
York City.) So now there are two record types: core geonames data and
alternates. The core records contain the main geonames data: IDs, canonical
name, country, admin divisions, etc., and they can be ingested by all clients
regardless of their locale or country. The alternates records are scoped by
language and are intended to be ingested only by clients with matching locales.

Improve geonames fetching and weather-suggestion matching so all admin levels
and countries are supported. e.g., "waterloo on", "waterloo canada", "waterloo
on canada", etc.

Relax the weather parsing a little to allow multiple weather keywords ("rain
weather").

Keep track of all available admin codes per geoname. There are four of them.
This is necessary because a lot of countries outside North America have multiple
admin levels, and determining whether a given geoname is related to another one
requires comparing their admin codes.

Instead of manually computing name variants and inserting them separately into
the DB, use a custom Sqlite collating sequence. ("Variants" here means removing
punctuation, lowercasing, removing diacritics, etc.)

Store each geoname's `ascii_name` as an alternate. That's useful for chars like
"ö", which is represented as "oe" in the ASCII name (at least the geonames data
I've seen).

Minor changes:

Store latitude and longitude and strings instead of floats. I made this change
to derive `Eq` for `Geoname`, but it makes sense anyway and is how I should have
done it originally.

Add `Geoname::geoname_type` so consumers can easily understand whether it's a
city, admin region, or country.

Remove the `geoname_type` param from `fetch_geonames`. Consumers can filter out
matching geonames that they don't want instead.

Author: 0c0w3

Cargo.lock

@@ -17,7 +17,7 @@ log = "0.4"
 once_cell = "1.5"
 parking_lot = ">=0.11,<=0.12"
 remote_settings = { path = "../remote_settings" }
-rusqlite = { version = "0.33.0", features = ["functions", "bundled", "load_extension"] }
+rusqlite = { version = "0.33.0", features = ["functions", "bundled", "load_extension", "collation"] }
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 error-support = { path = "../support/error" }
@@ -26,6 +26,9 @@ viaduct = { path = "../viaduct" }
 viaduct-reqwest = { path = "../support/viaduct-reqwest", optional=true }
 tempfile = { version = "3.2.0", optional = true }
 thiserror = "1"
+# This is an old version of `unicase` but it's the one mozilla-central uses.
+unicase = "2.6"
+unicode-normalization = "0.1"
 uniffi = { version = "0.29.0" }
 url = { version = "2.1", features = ["serde"] }
 
@@ -34,6 +37,7 @@ criterion = "0.5"
 env_logger = { version = "0.10", default-features = false }
 expect-test = "1.4"
 hex = "0.4"
+itertools = "0.14"
 rc_crypto = { path = "../support/rc_crypto" }
 
 [build-dependencies]