Cache flow_from_changes

[InnovativeInventor]

This is a PR broken out from #372 to ease the code review process. These sets of PRs are the second and final batch -- after every PR from this batch is merged in, the entire perf-tuning branch will be considered fully merged in and done.

[codecov-commenter]

Codecov Report

Merging #384 (faea096) into main (8acd44c) will decrease coverage by 0.01%.
The diff coverage is 94.28%.

@@            Coverage Diff             @@
##             main     #384      +/-   ##
==========================================
- Coverage   88.17%   88.16%   -0.02%     
==========================================
  Files          37       37              
  Lines        1666     1664       -2     
==========================================
- Hits         1469     1467       -2     
  Misses        197      197              

Impacted Files	Coverage Δ
gerrychain/grid.py	73.49% <0.00%> (ø)
gerrychain/proposals/tree_proposals.py	27.27% <50.00%> (ø)
gerrychain/constraints/contiguity.py	96.59% <100.00%> (ø)
gerrychain/metagraph.py	94.44% <100.00%> (ø)
gerrychain/partition/assignment.py	95.89% <100.00%> (-0.11%)	⬇️
gerrychain/partition/partition.py	98.76% <100.00%> (ø)
gerrychain/proposals/proposals.py	97.67% <100.00%> (ø)
gerrychain/proposals/spectral_proposals.py	87.50% <100.00%> (ø)
gerrychain/updaters/compactness.py	97.72% <100.00%> (ø)
gerrychain/updaters/county_splits.py	100.00% <100.00%> (ø)
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8acd44c...faea096. Read the comment docs.

[gabeschoenbach]

This all makes sense, my one question is in the new assignment of self.flows at line 81:

self.flows = flows_from_changes(parent, self)  # careful

self here refers to the partition, yes? It's okay to not be the first parameter? I see that when it's called in line 111 of tally.py you explicitly pass in a partition, so I guess I'm wondering why it necessarily needs to be called self in the first place.

[InnovativeInventor]

Dicts can't be hashed in Python (so it can't be cached if a flips object is passed, which was the original behavior). Partitions, on the other hand, can.