Patches for bugs introduced in 2.4 and 2.5 releases

[ianthetechie]

---

Here's the reason for this change 🚀

Versions 2.4.0 and 2.5.0 respectively introduced some issues, making certain kinds of input either unparseable or unsuitable for geocoding.

Demo links:

• https://parser.demo.geocode.earth/demo/#Calle%20Principal%2020%20Barcelona
• https://parser.demo.geocode.earth/demo/#main%20st%20seattl

---

Here's what actually got changed 👏

• Added a "not" restriction in one of the venue classification rules. Via reverse engineering, I determined that this was responsible for the "Main St Seattl" regression.
• Perhaps slightly more controversial, I removed a row from dictionaries/nb/concatenated_suffixes_separable.txt. This appears to be significantly interfering with Spanish addresses.

---

Here's how others can test the changes 👀

I've added unit tests for both cases. Additionally, I swapped in the branch for the pelias API and verified that all tests pass there. I actually discovered this while tinkering with some other things and realizing that like 100 API tests (slightly inflated in scope since it's char by char searches) fail as-is ☠️

I think longer term it'd probably be a good idea to harmonize that code and testing process a bit. Right now it's disjoint because the API has a second layer of processing logic on top of this lib, and that complicates testing, which is probably why these have been lurking.