Expand the benchmark suite to answer the performance questions raised by the latest undent work instead of relying on a few broad comparisons. The new cases measure short single-line string fast paths, already-clean inputs, mixed-newline inputs, Unicode-aware alignment, and scaling behavior across more realistic line counts. Add a small curiosity section for string-building strategies so repeat-heavy helpers can be compared against native primitives with the same harness. Keeping these measurements in their own commit makes it easier to review performance intent without mixing it into the user-facing feature diff.