Vertex/Arc ID sorting bug in Network.split_arcs()

[jGaboardi]

This PR addresses #526.

The were 2 sorting issues happening here.

1. The order of vertex IDs comprising an arc ID was causing issues during the split. When a network object is split some vertices are retained and some new vertices are added. This was causing issues with iterating over arcs.
2. Arc IDs were not being re-sorted following conversion back from a set after the network was split. When the arcs are fed into libpysal.weights.W it expects sorted IDs, this is what was causing all the counts all over the place.

[jGaboardi]

All tests are passing locally (macOS, Python 3.8) and passing CI for Ubuntu, Python 3.8, but then failing out on Ubuntu, Python 3.6. Appears to be an indexing/sorting discrepancy at first glance.

[jGaboardi]

Passing:

OS/Python	3.6	3.7	3.8
macOS	✅	✅	✅
Ubuntu	✅	✅	✅
Windows	❌	✅	✅

Python 3.6 on Windows failing due to an issue with fiona upstream. The issue started with fiona build win-64/fiona-1.8.17-py36hdef4c2b_0.tar.bz2 (see https://anaconda.org/conda-forge/fiona/files) and persists in win-64/fiona-1.8.17-py36hdef4c2b_1.tar.bz2.

[jGaboardi]

The set/indexing issue is resolved, but an issue with fiona upstream is causing failures on Windows builds.

[codecov]

Codecov Report

Merging #535 into master will increase coverage by 0.3%.
The diff coverage is 100.0%.

@@           Coverage Diff            @@
##           master    #535     +/-   ##
========================================
+ Coverage    97.6%   97.9%   +0.3%     
========================================
  Files           3       3             
  Lines        1136    1143      +7     
  Branches      284     286      +2     
========================================
+ Hits         1109    1119     +10     
+ Misses          6       4      -2     
+ Partials       21      20      -1     

Impacted Files	Coverage Δ
spaghetti/network.py	97.8% <100.0%> (+0.4%)	⬆️

[jGaboardi]

The current commit will fail, but we'll merge anyway in the hopes the fiona build issue with Python 3.6 on Windows will be resolved.