Consider inserting invisible groups, rather than parens, when applying grouping for precedence

[dtolnay]

As of #1659, we are currently using code like token::Paren::default().surround(tokens, |tokens| e.to_tokens(tokens)) .

But maybe a better behavior would be token::Group::default().surround(...).

[programmerjake]

I think using parens still is better because you may want to dump the tokens to a file, which breaks with invisible delimiters afaik