Clean up and document syntax tree child access API + mark public API

[c42f]

Here I commit to a more consistent but simpler child access API for syntax trees, as informed by the JuliaLowering work so far:

• is_leaf(node) is given a precise definition (previously !haschildren() - but that had issues - see Rename haschildren() to is_leaf() #483)
• children(node) returns the child list, or nothing if there are no children. The nothing might be seen as inconvenient, but mapping across the children of a leaf node is probably an error and one should probably branch on is_leaf first.
• numchildren(node) is documented
• node[i], node[i:j] are documented to index into the child list

We distinguish GreenNode and its implementation of span from SyntaxNode and its implementation of byte_range and sourcetext - these seem to just have very different APIs, at least as of now.

I've deleted the questionable overloads of multidimensional getindex and the child function in favor of single dimensional getindex. I don't know whether anyone ever ended up using these. But I didn't and they didn't seem useful+consistent enough to keep the complexity.

I've kept setindex! for now, to set a child of a SyntaxNode. Though I'm not sure this is a good idea to support by default and I didn't mark it public yet.

Also I've added more docs!

Also mark the public parts of the API as public in Julia 1.11+

[timholy]

This looks good. I think TypedSyntax's tests do exploit multidimensional child, but that's an easy change. I presume this will become 0.5.0?

[c42f]

I presume this will become 0.5.0?

We need to get to 1.0 so we can use semver properly. So the next release will be 1.0.0 :-) I've been working really hard on a lot of 1.0.0 changes in the last few weeks, trying to get the documentation and APIs in better shape. See also #472, but I keep finding more to do - it turns out there's a lot of changes I want to squeeze in which are informed by the JuliaLowering work.

[c42f]

I think TypedSyntax's tests do exploit multidimensional child, but that's an easy change.

We could keep child() if you think it's useful? In that case I'd document what it does and mark it public. It's more the multidimensional getindex which is dubious and should probably be deleted.

[timholy]

Either way is fine; it's not such a hard change to make to TypedSyntax's tests. A regex search suggests there are 74 uses of child(arg, i, j...) and that's not too bad to change if you think it would be cleaner without.

[c42f]

Ok thanks! Let's go with the more minimal API then :)