Allow :host and :host-context to take selectors…

[savetheclocktower]

…as well as :nth-child and :nth-last-child.

Fixes #57.

This was prompted by #57 and I wanted to fix it while it was still in my head. I got bitten by the other thing that this PR fixes…

:nth-last-child(-n + 3 of li.important) {}

…where li.important is parsed as a plain_value instead of as a selector.

This is another quick fix that avoids a larger reckoning with pseudoclass arguments. It's hard to say whether the CSS spec has any sort of predictable limitation in mind for what can be a valid argument to a pseudoclass; the ones we've whitelisted are those that take CSS selectors, but among the others it's generally a one-off set of valid values that varies per pseudoclass. For example:

• :state()’s argument is a custom state defined by a custom element — something we can't know, but which is probably constrained similarly to a JS identifier (begins with ASCII, no spaces, etc.)
• :lang()’s argument is a language or locale code (en, en-US)
• :dir()’s is either rtl or ltr
• :nth-of-type() has similar syntax to :nth-child, but without the optional filtering selector that the latter has — even, odd, 4, -n, 2n+1 are all valid values

In theory, we could get away with a minimal approach:

• Pseudoclass arguments that are obviously selectors get parsed as such
• All other text inside a pseudoclass’s parentheses is parsed as plain_value

In practice, we go a bit further than this; for instance, I think the 4 in :nth-child(4) would get parsed as an integer_value, which is fine.

As I describe in #57, the holy grail here would be to prevent plain_value from precluding selector detection while lexing, but I haven't figured that out yet. Until we do, the best approach is to manually handle each pseudoclass that can take a selector to prevent scenarios where both selector and plain_value are valid lexing options.

---

I forget what the right hygiene is these days around committing generated files, so let me know if I've done this wrong and I'll amend it.

[mohamedmansour]

@amaanq is it possible you can review this please! This unblocks Web Components support for tree-sitter-css since they depend on :host