levanter: log per-outlier encode time in BatchTokenizer

Wrap _encode_long_string in rigging.timing.log_time so wedged or
unexpectedly slow outliers surface in worker logs without needing
external profiling.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: ravwojdyla

lib/levanter/src/levanter/data/text/_batch_tokenizer.py

@@ -4,6 +4,7 @@
 from typing import Sequence, Any
 
 import regex
+from rigging.timing import log_time
 
 from levanter.data import BatchProcessor
 from levanter.tokenizers import MarinTokenizer
@@ -103,25 +104,26 @@ def _encode_long_string(self, text: str) -> list[int]:
         tokens, regardless of how long the original text is.
         """
         ids: list[int] = []
-        pieces: list[str] = []
-        remaining = text
-        while True:
-            if len(remaining) > self._workaround_len:
-                match = ws.search(remaining, self._workaround_len)
-                split = match.start() if match is not None else len(remaining)
-                pieces.append(remaining[:split])
-                remaining = remaining[split:]
-            else:
-                pieces.append(remaining)
-                remaining = ""
-
-            if len(pieces) >= _LONG_STRING_BATCH_SIZE or not remaining:
-                for encoded_piece in self.tokenizer.encode_batch(pieces, add_special_tokens=False):
-                    ids.extend(encoded_piece)
-                pieces.clear()
-
-            if not remaining:
-                break
+        with log_time(f"BatchTokenizer encoded {len(text):,}-char outlier record"):
+            pieces: list[str] = []
+            remaining = text
+            while True:
+                if len(remaining) > self._workaround_len:
+                    match = ws.search(remaining, self._workaround_len)
+                    split = match.start() if match is not None else len(remaining)
+                    pieces.append(remaining[:split])
+                    remaining = remaining[split:]
+                else:
+                    pieces.append(remaining)
+                    remaining = ""
+
+                if len(pieces) >= _LONG_STRING_BATCH_SIZE or not remaining:
+                    for encoded_piece in self.tokenizer.encode_batch(pieces, add_special_tokens=False):
+                        ids.extend(encoded_piece)
+                    pieces.clear()
+
+                if not remaining:
+                    break
 
         return ids