Incorporate feedback from Trevor/Alli

Specific tasks based on Trevor's feedback:

[x] #77
[x] #78
[x] Clarify that Table 1 represents the optimal cluster thresholds and their VI values based on training data and not the VI values for all datasets (resolved by 134f9ca)
[x] Clarify the purpose of or edit down the prominence of the early/late split for flu and SC2 (resolved by 4a3d3077)
[x] State clearly in the abstract and conclusion which single method we recommend for most uses
[x] Explicitly state in the discussion that automated labeling of reassorted or recombinant lineages is future work and outside the scope of this project (resolved by 86ac1a5f)
[x] #79
[x] Clarify reliability of genetic clusters from embeddings without relying entirely on expert clade definitions (resolved by 9bfbb951)
- Relationship between Euclidean/genetic distances shows how reliable embeddings are up to a certain genetic distance which is a key component of how clusters get defined
- Within/between group distances for most clusters match those of expert clades, suggesting that the diversity captured by clusters is comparable to what an expert would assign manually
- Cluster accuracy for each method depends in part on sampling density
- Cluster accuracy also depends on minimizing sampling bias
- Clusters tend to be monophyletic (quantify this)

Specific request from Nidia:

blab / cartography