Allow naked binaryOp at the start of a has selector

[richsilv]

Fixes #132, allowing users to write selectors with naked binary operators inside blocks, like FunctionExpression:has(> [name="foo"]).

This is achieved in two steps:

1. Update to the Peg.js grammar.
2. Handling the case of a missing left-hand selector inside a has block, by injecting a selector corresponding to the node which is then sub-traversed, using a new exactNode type.

[moniuch]

I would love this PR to get a review from the maintainers...

[michaelficarra]

This is too permissive. Instead, I would have just allowed an optional binaryOp in the has production.

[c-harding]

It can’t go directly in the has production, because that would cause problems if the second alternative has a root operator (:has(> a, > b)). I have resolved this in #145

[AustinGrey]

@richsilv Is there any appetite for addressing the reviewer's comments and resubmitting? This downstream affects ts-query which I need for another project, and I unfortunately do not know pegjs to try and submit my own PR.