remark: Fenced code block lang and meta not separated by curly bracket

Subject of the issue

When splitting the code block flag into lang and meta, remark does not account for this case:

```js{foo=1 bar=2 baz=3}
console.log("hello world");
```

According to the current implementation, only spaces and tabs are considered as separators:

https://github.com/remarkjs/remark/blob/8d8e3bcfbc1c89b8eae9119703f13414269ebf73/packages/remark-parse/lib/tokenize/code-fenced.js#L230-L245

Not taking { into the account leads to an issue in Prettier. For example, a flag like mylang{foo="hello   world"} is split into mylang{foo="hello and world"} instead of mylang and {foo="hello   world"}. When Prettier stitches back lang and meta, the result becomes mylang{foo="hello world"}, i.e. extra spaces are lost. See Playground.

If accepting { as a separator between lang and meta can count as a bugfix, I’m happy to open a PR that would add it. Doing so will save Prettier from a need to hack the lang value – this workaround dates back to remark v5, i.e. before #345.

Looking at how Prettier does const langMatch = node.lang.match(/^[\w-]+/);, perhaps we can also do the same in remark. I can’t imagine a scenario in which lang would not match this regex 🤔

Your environment

• OS: web
• Packages: remark-parse v8.0.3 via prettier v2.1.1
• Env: browser