Implement sigma-separation

[cthoyt]

Closes #120

This PR adds a high-level interface and implements tests for sigma-separation, a generalization of d-separation that works not only for directed acyclic graphs, but also for directed graphs containing cycles. It was originally introduced in

Constraint-based Causal Discovery for Non-Linear Structural Causal Models with Cycles and Latent Confounders
Forré and Mooij. 2019. arXiv:1807.03024

and is an integral part of cyclic ID algorithm (see #71) and the gID algorithm (see #72)

References/Notes

• https://stats.stackexchange.com/questions/586810/sigma-separation-question-in-cyclic-causal-graph-understanding-sigma-separatio
• Author's implementation: https://github.com/caus-am/sigmasep

[djinnome]

Once $\sigma$-separation is implemented, then there is also the cyclic ID algorithm (#71):

Causal Calculus in the Presence of Cycles, Latent Confounders and Selection Bias.
Forré, P., & Mooij, J. (2019). arXiv:1901.00433.

[codecov]

Codecov Report

Merging #150 (1110763) into main (d676831) will increase coverage by 0.20%.
The diff coverage is 91.30%.

@@            Coverage Diff             @@
##             main     #150      +/-   ##
==========================================
+ Coverage   79.85%   80.06%   +0.20%     
==========================================
  Files          39       40       +1     
  Lines        2477     2523      +46     
  Branches      566      576      +10     
==========================================
+ Hits         1978     2020      +42     
- Misses        412      414       +2     
- Partials       87       89       +2     

Files Changed	Coverage Δ
src/y0/algorithm/sigma_separation.py	91.30% <91.30%> (ø)