Drop quotes from names parsed from fea earlier so we detect empty ones correctly

[rsheeter]

Parsing NameSpec from fea retains the double quotes, dropping them only when generating final name records. That causes fontc to produce unwanted empty name records because it filters using is_empty. Naive solution: drop the quotes earlier so is_empty works as expected.

Makes name identical for python resources/scripts/ttx_diff.py 'https://github.com/marcologous/hanken-grotesk#sources/HankenGrotesk.glyphs'.

[cmyr]

so I'm still slightly confused about where the is_empty check is, but if I'm understanding the intent of this patch correctly, I think that the cleanest fix is to replace the string() method on the NameSpec type in token_tree::typed with something like,

    pub(crate) fn string(&self) -> &str {
        self.find_token(Kind::String)
            .unwrap()
            .text
            .trim_matches('"')
    }

(although honestly just doing it inline in resolve_name_spec seems fine too.)

...and then since the actual bug is with something ending up in the binary, the best way to test this would be to add a new case in fea-rs/test-data/compile-tests; there's a readme there that explains this process but it's easy.

BUT if part of the complication here is that there is some filtering happening outside of fea-rs then that won't work, and in that case we should figure out if fea-rs should be doing this filtering or not.

I've left a few little comments on the PR as written, for informational purposes.

(i'm not marking this approved yet because I want to understand where the filtering you mention is happening first)

[cmyr]

trivia, but fwiw you don't need the vec here you can just have these be arrays..

[cmyr]

we generate a node() method you can call, so this shouldn't be necessary.

[rsheeter]

This was to enable construction of a test value, not reading, but perhaps that will be unnecessary once revised

[cmyr]

nice okay ya I do think this is much cleaner. Thanks!