fix parsing errors about template literals (fixes #575)

[mysticatea]

Fixes #575.

In src/tokencontext.js#L112-L119, class and function keywords push new context into this.context. However, there is no chance to pop the context from this.context if the keyword was consumed as an identifier such as {class: 1}. This fact breaks the tokenization of template literals.

This PR makes popping the context when a class/function keyword is consumed as an identifier.

I'm not sure whether this fixing is correct or not. Because this bug is in tokenization process, but this PR modifies parsing process. Please tell me (or feel free to fix it yourself) if the more proper way exists.

[marijnh]

Thanks, I've merged this. It's technically a bit problematic, because it breaks the separation between the tokenizer and the parser, and thus won't help tokenize such programs, but I can't think of a good way to distinguish identifiers in object literal property or class method position during tokenizing, so we'll just have to live with that for now.

[mysticatea]

@marijnh could you make a new release?

[marijnh]

Sure. I've released 5.2.0