- Explicity split entry string by space during split by max_tokens
- Prevent formatting of compiled entry from being lost
- The formatting itself contains useful information
No point in dropping the formatting unnecessarily,
even if (say) the currrent search models don't account for it (yet)