Update networkx version

[eccabay]

Closes #3972

Networkx changed the topological order of their DAGs with version 2.6, which broke some of our tests that validate the computation order. We pinned it and left it there. Changes here unpin the version and update the tests to be more flexible about the order of computations, so long as the data flow is still maintained.

[codecov]

Codecov Report

Merging #4035 (99f39b2) into main (368384c) will decrease coverage by 15.3%.
The diff coverage is 80.6%.

@@           Coverage Diff            @@
##            main   #4035      +/-   ##
========================================
- Coverage   99.7%   84.4%   -15.3%     
========================================
  Files        349     349              
  Lines      37242   37257      +15     
========================================
- Hits       37121   31425    -5696     
- Misses       121    5832    +5711     

Impacted Files	Coverage Δ
...valml/tests/automl_tests/test_default_algorithm.py	0.0% <0.0%> (-100.0%)	⬇️
...valml/tests/pipeline_tests/test_component_graph.py	99.9% <100.0%> (+0.1%)	⬆️
evalml/tests/pipeline_tests/test_pipelines.py	99.9% <100.0%> (ø)
evalml/tests/automl_tests/test_search.py	0.0% <0.0%> (-100.0%)	⬇️
evalml/tests/automl_tests/test_progress.py	0.0% <0.0%> (-100.0%)	⬇️
evalml/tests/tuner_tests/test_skopt_tuner.py	0.0% <0.0%> (-100.0%)	⬇️
evalml/tests/automl_tests/test_engine_base.py	0.0% <0.0%> (-100.0%)	⬇️
evalml/tests/automl_tests/test_automl_utils.py	0.0% <0.0%> (-100.0%)	⬇️
evalml/tests/automl_tests/test_search_iterative.py	0.0% <0.0%> (-100.0%)	⬇️
evalml/tests/tuner_tests/test_grid_search_tuner.py	0.0% <0.0%> (-100.0%)	⬇️
... and 27 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[jeremyliweishih]

is the component ordering changing? Just wondering why we're changing these tests

[eccabay]

Yep, that was the change that caused us to pin the version a year and a half ago. For non-linear graphs, the order is now more breadth-first rather than depth-first. It's still a correct ordering, which is why the tests now simply assert that components that must be fit/transformed before others still are.

[jeremyliweishih]

Thanks!