Simplify error handling a bit

[vihu]

Summary

Rather than crashing because of an expect statement in the code, I have added thiserror library and simplified a bit of error handling so the callers can handle it a bit more gracefully.

NOTE: There's more occurrences of expect statement being used throughout. I have only replaced one which I found happening more often in my project. A more thorough handling of errors is probably wiser in the long term.

[tom-whitehead]

Hey @vihu, Unfortunately this is a breaking change requiring a major version bump, so I would prefer to avoid adding a new error type as I think people would be matching on it. Unfortunately, I didn't know about #[non_exhaustive] when I wrote this! I am building up a list of things I want to change when I release a new major version.

The reason for all of the expects is partially because I suck at Rust 😅, but also partially because I thought they were all theoretically unreachable. In the most part, I'm expecting a node id to be in the tree of nodes from which I took it in the first place. This should be true and if it's not it's more likely that it's a bug. Do you have a toy dataset or a test case I can reproduce it with? (sorry will also try and look at your other PR again later)

[tom-whitehead]

@vihu I just saw the test from your other PR hits this panic. When there's only one cluster and it's the single root cluster I need to skip the epsilon check. I will separately raise a PR for that. I think this should only be hit for very small datasets when allow_single_cluster is true. It looks like fixing this will fix the test in your other PR too.
Thanks for unearthing all these bugs.

[tom-whitehead]

@vihu this is the PR if you want to try it out.