Recheck project filenames and Word commenters

src/lib/project-review.ts

@@ -9,6 +9,20 @@ export type BatchItem = { key: string; text: string };
 export type BatchSegment = BatchItem & { start: number; end: number };
 export type TextBatch = { text: string; segments: BatchSegment[] };
 
+/** Filename separators look like spaces to NER, except inside known literal terms. */
+export function filenameScanText(stem: string, knownTerms: Iterable<string>): string {
+  const protectedOffsets = new Set<number>();
+  for (const term of knownTerms) {
+    if (!term.includes("_")) continue;
+    const escaped = term.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+    const pattern = new RegExp(`(?<![\\p{L}\\p{N}])${escaped}(?![\\p{L}\\p{N}])`, "giu");
+    for (const match of stem.matchAll(pattern)) {
+      for (let offset = match.index; offset < match.index + match[0].length; offset++) protectedOffsets.add(offset);
+    }
+  }
+  return stem.replace(/_/g, (_character, offset: number) => protectedOffsets.has(offset) ? "_" : " ");
+}
+
 /** Keep file boundaries while reducing many small text/path API calls to a few batches. */
 export function makeTextBatches(items: BatchItem[], limit = MAX_BATCH_CODEPOINTS): TextBatch[] {
   const batches: TextBatch[] = [];