parse, lex: Decompose AST extraction and forward-declare dump types (#7645)

- Enhance toolchain/parse/node_kind.def with category x-macros
(CARBON_PARSE_NODE_KIND_DECLARATION, CARBON_PARSE_NODE_KIND_EXPRESSION,
CARBON_PARSE_NODE_KIND_PATTERN, CARBON_PARSE_NODE_KIND_STATEMENT) that
default to CARBON_PARSE_NODE_KIND, keeping node_kind.def as the single
source of truth without manual expansion divergence.
- Decompose the monolithic toolchain/parse/extract.cpp translation unit
into separate compilation units for declarations, expressions, patterns,
and statements by expanding respective category x-macros.
- Forward-declare Tree and TokenizedBuffer in parse and lex dump headers
to reduce header inclusion depth.

Assisted-by: Antigravity with Gemini

toolchain/parse/extract_patterns.cpp

@@ -0,0 +1,17 @@
+// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
+// Exceptions. See /LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include "toolchain/parse/extract.h"
+
+namespace Carbon::Parse {
+
+#define CARBON_PARSE_NODE_KIND(KindName)
+#define CARBON_PARSE_NODE_KIND_PATTERN(KindName)                       \
+  template auto TreeAndSubtrees::TryExtractNodeFromChildren<KindName>( \
+      NodeId node_id,                                                  \
+      llvm::iterator_range<TreeAndSubtrees::SiblingIterator> children, \
+      ErrorBuilder * trace) const -> std::optional<KindName>;
+#include "toolchain/parse/node_kind.def"
+
+}  // namespace Carbon::Parse