make EdgeLabel required in the CST

[OmarTawfik]

Previously, some nodes didn't produce labels, so Edge::label was Option<>. This is no longer the case.
Making this a required property for all edges (replacing the Option) makes it much easier to handle/match in client code.

[AntonyBlakey]

The ability to have un-named nodes is critical to the query system!

[Xanewok]

I'm also in the camp of having the names optional. If it's all about Option<T> handling, we can surely inline the None variant directly in the FieldName but it makes sense not to name every node there is.

[OmarTawfik]

I really think we should make it required everywhere. From an API perspective, if i'm iterating on children, or even filtering/inspecting a specific child, I should be able to read/know its name, and it should be strongly typed, and finite/matchable. All such is provided by the FieldName enum. But using Option<FieldName> and None force me to then rely on my own knowledge of how the parser works to know what to do next.

Re: the query system: I think the fact that something has a name or not shouldn't interfere with the querying system, as they are independent, and can actually be confusing in some cases, like the separator: Comma that separates TupleDeconstructionElements with empty members. I think it should derive its contentful-ness from other existing properties accordingly.

But we also discussed the cost of adding a side channel to record such information, and it is not clear whether it is worth doing now.

[OmarTawfik]

Another idea: we might not need a side channel after all, if we can derive this info from existing node kinds (example below). We would need to use #808 to see if this affects query engine performance:

fn is_node_contentful(parent_kind: &RuleKind, child_kind: &TokenKind) -> bool {
    match (parent_kind, child_kind) {
        // for `StructItem`, add children that appear in `terminated_by`/`delimited_by`:
        (RuleKind::Statement, TokenKind::Semicolon) => false,
        (RuleKind::Block, TokenKind::LeftCurly | TokenKind::RightCurly) => false,
        // .....

        // For `Separated`, add the `separator`:
        (RuleKind::Arguments, TokenKind::Comma) => false,
        (RuleKind::Parameters, TokenKind::Comma) => false,
        // .....

        // Also trivia:
        (_, TokenKind::Whitespace | TokenKind::Newline | TokenKind::Comment) => false,

        // otherwise, contentful:
        _ => true,
    }
}

[OmarTawfik]

Let's defer this until we have more experience/feedback on the API.

[OmarTawfik]

@AntonyBlakey Currently we already produce labels for all (?) children nodes in the tree. The remaining work is about changing the type of the Edge::label field from Option<EdgeLabel> to EdgeLabel.
It shouldn’t impact the query engine or the parser, as users should still be able to author queries with no edge filters. But it will make it much easier to use our CST and other APIs, and eliminate a lot of redundant code. Please let me know if you have concerns on this.