Decide whether `K"parens"` should be visible to `SyntaxNode`

[c42f]

As of #222 etc we have a parse node for grouping parentheses. This is extremely useful for tooling and has allowed several parsing difficulties to be solved neatly.

But it's pretty inconvenient when using SyntaxNode in a way that few other tree nodes are - parentheses are permitted in many unusual situations in the reference parser, and often treated as completely invisible to the parser.

Also SyntaxNode is AST-like, but grouping parens are arguably syntax trivia and shouldn't be in the AST anyway.

So maybe we should include parens only in the green tree, but elide them when converting to SyntaxNode.

[c42f]

Ugh this is pretty awkward no matter which way you look at it. Diverging the GreenNode and SyntaxNode trees for this one particular case feels bad.

Also when I tried to fix this in #265 I found that approach (make SyntaxNode point to its outermost GreenNode parent of type `K"parens") quite awkward as described there.

Another alternative approach is to have SyntaxNode always point to the associated GreenNode of the same kind. In this case, no SyntaxNode would ever point to a GreenNode of type K"parens".

The irritation there is that we occasionally need in Expr conversion to know whether a SyntaxNode was wrapped in a GreenNode of type K"parens" for emulating some of the Expr inconsistencies.

Hmm. Maybe it's time to convert directly from GreenNode to Expr, avoiding SyntaxNode entirely? That would be more efficient in any case...