Update html5lib-tests

[untitaker]

Follow-up items:

• More precise errors
• There is still large divergence between some of the states and their names in code vs spec. It's just that there's no testcase for them.

I'm also fuzzing this branch continuously against https://github.com/untitaker/html5gum using FUZZ_HTML5EVER=1 to find more bugs, but they only cover the tokenizer. Some bugfixes have been made and corresponding regression tests added in a regression.test file.

Fixes #459.

[jdm]

What makes this a WIP?

[untitaker]

Hmm, in the sense that it's not mergeable as-is. I hadn't planned on working it for now though

[untitaker]

eh, I'm working on this now.

• Some of the testsuite format has changed. Errors are no longer tokens, they're their own separate array entry with line/col/code.
• Ignoring line/col, the code does not match your human-readable errors at all. Do we want to make an effort assimilating those to what html5lib-tests prescribes?

that's all there is to tokenizer tests, haven't looked at treebuilder tests yet. That said I am not really familiar with how treebuilding is done in the spec, so this could take quite a while.

[jdm]

Good question about error codes. I suspect a decent way to handle that would be to make an actual error enum that gets passed to emit_error, and then make the html5lib test harness map those error variants to the codes that the html5lib tests expect.

That being said, I'm inclined to file an issue about that and ignore specific error content for right now.

[untitaker]

@jdm all tests are fixed. can you take a look?

[EdorianDark]

@mrobinson Can you take a look maybe?