Remove most of the metaprogramming in node.h in favor of listing all the members in the typed node structs. (#3310)

Split `node.h` into separate files for ID types (`id.h`) and for typed
nodes (`typed_nodes.h`). The per-node-kind data is now specified as part
of declaring the typed nodes, and is removed from the node kinds
x-macros, which now simply enumerate the node kinds.

toolchain/check/handle_variable.cpp

@@ -41,7 +41,7 @@ auto HandleVariableDeclaration(Context& context, Parse::Node parse_node)
     init_id = Initialize(context, parse_node, var_id, init_id);
     // TODO: Consider using different node kinds for assignment versus
     // initialization.
-    context.AddNode(SemIR::Assign(parse_node, var_id, init_id));
+    context.AddNode(SemIR::Assign{parse_node, var_id, init_id});
   }
 
   context.node_stack()