perf(parser): cache syntax errors to skip walk on cache hit

extractErrors is a pure function of the parsed tree (it ignores content), so re-walking the whole tree on every cache hit wasted work on the hottest path (file_read+include_ast, ast_query, edit validation). Store []SyntaxError in CachedTree and return it directly on hits. Adds TestParseCachedErrorsMatch asserting cache-hit errors equal the cache-miss parse.

internal/parser/parser.go

@@ -48,10 +48,13 @@ type Registry struct {
 }
 
 // CachedTree stores a parsed tree with its metadata.
-// Content is not stored to reduce memory usage.
+// Content is not stored to reduce memory usage. Syntax errors are a pure
+// function of the tree (extractErrors ignores content), so they are cached
+// alongside it to avoid re-walking the whole tree on every cache hit.
 type CachedTree struct {
 	Tree     *sitter.Tree
 	Language protocol.Language
+	Errors   []SyntaxError
 }
 
 // ParseResult contains the result of parsing a file.
@@ -202,11 +205,10 @@ func (r *Registry) Parse(ctx context.Context, filename string, content []byte) (
 	hash := fmt.Sprintf("%s:%016x", string(lang), xxhash.Sum64(content))
 	if cached, ok := r.cache.Get(hash); ok && cached.Language == lang {
 		r.cacheHits.Add(1)
-		errors := extractErrors(cached.Tree.RootNode(), content)
 		return &ParseResult{
 			Tree:     cached.Tree,
 			Language: lang,
-			Errors:   errors,
+			Errors:   cached.Errors,
 			Content:  content,
 		}, nil
 	}
@@ -242,10 +244,12 @@ func (r *Registry) Parse(ctx context.Context, filename string, content []byte) (
 	// Extract syntax errors
 	errors := extractErrors(tree.RootNode(), content)
 
-	// Cache result (LRU cache handles eviction automatically)
+	// Cache result (LRU cache handles eviction automatically).
+	// Errors are cached so cache hits skip a full tree walk.
 	r.cache.Add(hash, &CachedTree{
 		Tree:     tree,
 		Language: lang,
+		Errors:   errors,
 	})
 
 	return &ParseResult{

internal/parser/parser_test.go

@@ -172,6 +172,43 @@ func TestParseWithSyntaxErrors(t *testing.T) {
 	}
 }
 
+// TestParseCachedErrorsMatch verifies that the syntax errors returned on a
+// cache hit are identical to those computed on the initial parse. Errors are
+// cached alongside the tree (they are a pure function of the tree), so the hot
+// cache-hit path must not lose or alter them.
+func TestParseCachedErrorsMatch(t *testing.T) {
+	r := NewRegistry()
+	defer r.Close()
+
+	content := []byte("package main\n\nfunc main( {}\n") // missing closing paren
+	ctx := context.Background()
+
+	first, err := r.Parse(ctx, "test.go", content)
+	if err != nil {
+		t.Fatalf("first parse failed: %v", err)
+	}
+	if len(first.Errors) == 0 {
+		t.Fatal("expected syntax errors on first parse")
+	}
+
+	second, err := r.Parse(ctx, "test.go", content)
+	if err != nil {
+		t.Fatalf("cached parse failed: %v", err)
+	}
+	if second.Tree != first.Tree {
+		t.Fatal("expected cache hit (same tree pointer)")
+	}
+
+	if len(second.Errors) != len(first.Errors) {
+		t.Fatalf("cached errors count = %d, want %d", len(second.Errors), len(first.Errors))
+	}
+	for i := range first.Errors {
+		if second.Errors[i] != first.Errors[i] {
+			t.Errorf("cached error[%d] = %+v, want %+v", i, second.Errors[i], first.Errors[i])
+		}
+	}
+}
+
 func TestIsBinary(t *testing.T) {
 	tests := []struct {
 		name    string