improved string normalization for deduplication

[missinglink]

this PR improves the normalization function used for deduplication:

• remove diacritics
• improve test coverage

[missinglink]

I staged this to take a look and there were three notable changes:

Aggregate test results
Pass: 610
Improvements: 1
Fail: 115
Placeholders: 0
Regressions: 2
Took 114607ms
Test success rate 99.72%

I had a look and it seems as though they are actually positive but marked as failing due to brittle test cases.

General Placeholder requests
  ✘ regression [17] "{"text":"Schöneberg, DE"}": score 2 out of 3
  diff:
    gid
      expected: whosonfirst:locality:101907543
      actual:   whosonfirst:locality:1126037037

search city/country
 ✔ improvement [15] "{"text":"Akmenė, lithuania"}"

Who's on First regions
  ✘ regression [8] "{"text":"Xaçmaz, Azerbaijan","sources":"wof"}": score 3 out of 5
  diff:
    layer
      expected: region
      actual:   locality
    name
      expected: Xacmaz
      actual:   Xaçmaz

[orangejulius]

Looks good to me, is there anything else to do here before merging?

[Joxit]

LGTM

[missinglink]

I fixed some test issues and also now use our fancy new unicode.normalize() method to fix any potential unicode issues 🎉

This is good to merge once the failing test cases I mentioned above have been investigated.

[missinglink]

force pushed to rebase origin/master

[missinglink]

before

Aggregate test results
Pass: 615
Improvements: 10
Fail: 123

after

Aggregate test results
Pass: 645
Improvements: 5
Fail: 128

---

There is one notable test failure but I switched to dev->api to the same hash as prod and it's there too, so it's not the fault of this work 👍

search focus.point
  ✘ regression [1] "/v1/search?focus.point.lat=42.64&focus.point.lon=-84.7&text=10010": score 3 out of 4
  diff:
    priorityThresh is 1 but found at position 2

https://pelias.github.io/compare/#/v1/search?text=10010

[missinglink]

The three failing test cases mentioned in the previous comment are no longer failing.