Add tests for TOC generation

[shonfeder]

Followup to #240

• Testing it properly
• Add tests
• Add support for it to the CLI

As per #240 (comment)

[ghost]

This might be helpful: remark toc tests.

[sonologico]

@shonfeder, I'm thinking of adding support for different types of tests by having different labels at the end of the lines separating the test cases. What do you think of this approach?

```````````````````````````````` example
...
.
...
````````````````````````````````

```````````````````````````````` example-toc
...
.
...
````````````````````````````````

[shonfeder]

@sonologico sorry for the delay!

Would your approach mean altering the text of the current spec.txt used for testing?

If not, I wonder if could just different file names for different kinds of tests?

More generally, I'm wondering if we shouldn't split additional tests into two groups:

1. Integration tests, that would either use mdx or the cram-style tests used in dune.
2. Unit tests, that just confirm the expected AST is generated, or that a fragment renders as expected (for which we could use e.q. like ppx_expect if we want crams-style testing).

I don't have a hard opinion here, except that I find the current testing scheme a bit cumbersome to work with, and only justified by the fact that we are presumably working from the original spec text, which is itself left untouched.

[sonologico]

I was planning on adding more another test files. We already have more besides spec.txt, but actually, I think it's more sensible to have additional tests (besides the spec) using unit tests as you're suggesting. Which testing framework would you prefer? I don't have a preference. I have used alcotest and ounit in the past. Haven't used ppx_expect yet, but I'm open to it.

[shonfeder]

Ok! I'm happy for you to proceed however you see fit, tho also happy to offer opinions where useful :) If you do end up deciding the current testing approach is sufficient and preferable, and worth extending, I wouldn't begrudge the effort.

I in my own projects, I've preferred alcotest over ounit, and qcheck (sometimes on top of alcotest) over all.

ppx_expect would be especially useful if we want to give inline tests that illustrate behavior. As an example:

https://github.com/shonfeder/um-abt/blob/8c2a205ea1cc143b97dfaac7312500ffd93718e2/lib/abt.ml#L286-L359

It would be convenient for cases where we want to actually see the generated HTML, but aren't testing to a pre-given spec.

[sonologico]

Played with the idea here: https://github.com/ocaml/omd/blob/sonologico/toc-ppx_expect/src/omd.ml#L29.
I can see this being nice for some lightweight testing.

I think we'll end up needing something like alcotest, so I'm not sure if it's worth having both, but ppx_expect seems to be really nice work with.