fix(chunkers): also strip named capture groups in regex patterns

Named groups (?<name>...) are still capturing groups so split() interleaves
their matched text. Convert them to non-capturing alongside plain ( groups.

Author: waleedlatif1

apps/sim/lib/chunkers/regex-chunker.test.ts

@@ -206,6 +206,27 @@ describe('RegexChunker', () => {
       }
     )
 
+    it.concurrent(
+      'should not include delimiter text when pattern uses named capture groups',
+      async () => {
+        const chunker = new RegexChunker({
+          pattern: '(?<sep>---)',
+          chunkSize: 1024,
+          strictBoundaries: true,
+        })
+        const text = 'Section one content.---Section two content.---Section three content.'
+        const chunks = await chunker.chunk(text)
+
+        expect(chunks).toHaveLength(3)
+        expect(chunks[0].text).toBe('Section one content.')
+        expect(chunks[1].text).toBe('Section two content.')
+        expect(chunks[2].text).toBe('Section three content.')
+        for (const chunk of chunks) {
+          expect(chunk.text).not.toBe('---')
+        }
+      }
+    )
+
     it.concurrent('should leave non-capturing groups and lookarounds intact', async () => {
       const chunker = new RegexChunker({
         pattern: '(?=\\n\\s*\\{\\s*"id"\\s*:)',

apps/sim/lib/chunkers/regex-chunker.ts

@@ -15,11 +15,14 @@ const logger = createLogger('RegexChunker')
 
 const MAX_PATTERN_LENGTH = 500
 
+const NAMED_GROUP_PREFIX = /^\(\?<[^>]+>/
+
 /**
- * Converts unescaped capturing groups `(...)` into non-capturing groups `(?:...)`.
- * `String.prototype.split()` interleaves captured groups into the result array,
- * which would surface delimiter text as spurious chunks. Lookarounds, named
- * groups, and other `(?...)` constructs are left untouched.
+ * Converts unescaped capturing groups `(...)` and named capturing groups
+ * `(?<name>...)` into non-capturing groups `(?:...)`. `String.prototype.split()`
+ * interleaves captured text (named or otherwise) into the result array, which
+ * would surface delimiter text as spurious chunks. Lookarounds (`(?=`, `(?!`,
+ * `(?<=`, `(?<!`) and other `(?...)` constructs are left untouched.
  */
 function toNonCapturing(pattern: string): string {
   let result = ''
@@ -33,9 +36,17 @@ function toNonCapturing(pattern: string): string {
     }
     if (c === '[') inClass = true
     else if (c === ']') inClass = false
-    if (!inClass && c === '(' && pattern[i + 1] !== '?') {
-      result += '(?:'
-      continue
+    if (!inClass && c === '(') {
+      if (pattern[i + 1] !== '?') {
+        result += '(?:'
+        continue
+      }
+      const namedMatch = pattern.slice(i).match(NAMED_GROUP_PREFIX)
+      if (namedMatch) {
+        result += '(?:'
+        i += namedMatch[0].length - 1
+        continue
+      }
     }
     result += c
   }