Move child count and bracketing information for parse nodes into the node kind definition. (#4000)

Instead of tracking the bracketing and child count information in the
kind macro in `node_kind.def`, provide it to `NodeKind::Define` in
`typed_nodes.h`. If a node is both bracketed and has a fixed child
count, track both facts and check them both in tree verification, since
it's easy to do so now.

The overall goal here is to reduce `node_kind.def` down to a simple list
of names. I have a slightly different approach in mind for the token
kinds.

---------

Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
Co-authored-by: Carbon Infra Bot <carbon-external-infra@google.com>

toolchain/parse/tree.cpp

@@ -211,6 +211,7 @@ auto Tree::Verify() const -> ErrorOr<Success> {
 
     int subtree_size = 1;
     if (n_impl.kind.has_bracket()) {
+      int child_count = 0;
       while (true) {
         if (nodes.empty()) {
           return Error(
@@ -220,7 +221,17 @@ auto Tree::Verify() const -> ErrorOr<Success> {
         }
         auto child_impl = node_impls_[nodes.pop_back_val().index];
         subtree_size += child_impl.subtree_size;
+        ++child_count;
         if (n_impl.kind.bracket() == child_impl.kind) {
+          // If there's a bracketing node and a child count, verify the child
+          // count too.
+          if (n_impl.kind.has_child_count() &&
+              child_count != n_impl.kind.child_count()) {
+            return Error(llvm::formatv(
+                "NodeId #{0} is a {1} with child_count {2}, but encountered "
+                "{3} nodes before we reached the bracketing node.",
+                n, n_impl.kind, n_impl.kind.child_count(), child_count));
+          }
           break;
         }
       }