Improvement: Bound networkx to less than 3.3

[andrew-causalens]

Description and Motivation

networkx 3.3 (release notes) claims to fix a bug (PR) on simple paths but it actually returns paths of length one when source and destination is the same. This is suboptimal behavior particularly if there is no self loop edge.

Additional Context

We will add 3.3 support in future but first need to understand impact on our other packages. Plus our base image for 2024 Q2 is using networkx 3.2.1.

How Has This Been Tested?

Ensure we are using networkx 3.2.1 in unit tests for python 3.11. The actions on this PR do that.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Improvement (non-breaking change which improves functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Refactor (moving code around, general refactoring improvements)
• Documentation (documentation)

PR Checklist:

• I have reviewed my own PR? (review the PR as you are reviewing someone else's PR)
• I have implemented all requirements? (see JIRA, project documentation)
• I am affecting someone else's work? If so: I have added those people as reviewers?
• I have added comments to all the bits that are hard to follow
• At a bare minimum, I tested this manually
• I have added unit test(s)
• Is this a bugfix? If so have I added a regression test?
• Does this PR satisfy the one PR, one concern rule?
• If needed, documentation has been added/updated?
• I have updated the appropriate changelog with a line for my changes
• If this PR breaks reproducibility, have I added a note in the changelog explaining the break in reproducibility
  and its extent?

Screenshots (if appropriate):

[maxelliott-causalens]

changelog?

[maxelliott-causalens]

🙏