Re-parse with lint.NewParser for link-ref lookup

hasTOCLinkReference previously used goldmark.New(), which omits
mdsmith's PIBlockParserPrioritized. Extract the parser construction in
lint.NewFile into an exported lint.NewParser() so rules that need a
secondary parse (here, to consult the link reference map for [TOC]
suppression) stay consistent with the original lint parse instead of
drifting from it. No behavior change on the current test corpus;
defensive alignment for future parser additions.

Addresses review feedback on PR #144.

Author: claude

internal/lint/file.go

@@ -55,10 +55,13 @@ func (f *File) GetGitignore() *GitignoreMatcher {
 	return f.gitignoreVal
 }
 
-// NewFile parses source as Markdown and returns a File.
-func NewFile(path string, source []byte) (*File, error) {
-	reader := text.NewReader(source)
-	p := parser.NewParser(
+// NewParser returns a goldmark parser configured identically to the one
+// used by NewFile. Rules that need to re-inspect a document (for example,
+// to consult the link reference definition map) should use this so that
+// processing-instruction blocks and other mdsmith-specific parsing
+// decisions stay consistent with the original lint parse.
+func NewParser() parser.Parser {
+	return parser.NewParser(
 		parser.WithBlockParsers(
 			append(parser.DefaultBlockParsers(),
 				PIBlockParserPrioritized(),
@@ -71,7 +74,12 @@ func NewFile(path string, source []byte) (*File, error) {
 			parser.DefaultParagraphTransformers()...,
 		),
 	)
-	node := p.Parse(reader)
+}
+
+// NewFile parses source as Markdown and returns a File.
+func NewFile(path string, source []byte) (*File, error) {
+	reader := text.NewReader(source)
+	node := NewParser().Parse(reader)
 
 	lines := bytes.Split(source, []byte("\n"))
 

internal/rules/tocdirective/rule.go

@@ -10,7 +10,6 @@ import (
 
 	"github.com/jeduden/mdsmith/internal/lint"
 	"github.com/jeduden/mdsmith/internal/rule"
-	"github.com/yuin/goldmark"
 	"github.com/yuin/goldmark/ast"
 	"github.com/yuin/goldmark/parser"
 	"github.com/yuin/goldmark/text"
@@ -120,17 +119,17 @@ func buildMessage(token string) string {
 }
 
 // hasTOCLinkReference returns true when the document defines a link
-// reference with label "TOC" (CommonMark-normalized). Re-parsing is the
-// simplest way to delegate label normalization and code-block scoping to
-// goldmark itself, rather than approximating them with a source-level
-// scan.
+// reference with label "TOC" (CommonMark-normalized). It re-parses with
+// lint.NewParser so the parser configuration (including mdsmith's PI
+// block parser) matches the original lint parse; otherwise content
+// absorbed into a processing-instruction block could register as a link
+// reference here while being hidden from the rule's AST walk.
 func hasTOCLinkReference(source []byte) bool {
 	if len(source) == 0 {
 		return false
 	}
-	md := goldmark.New()
 	ctx := parser.NewContext()
-	md.Parser().Parse(text.NewReader(source), parser.WithContext(ctx))
+	lint.NewParser().Parse(text.NewReader(source), parser.WithContext(ctx))
 	_, ok := ctx.Reference("toc")
 	return ok
 }