scripts/freeze_paper_results.py's generated MANIFEST.md and docstring still say "what a paper should cite," contradicting the lifted freeze

Where: scripts/freeze_paper_results.py:1 (module docstring: "Freeze the current results/ into paper/results-frozen/ for a paper citation") and line 110 (hardcoded into every generated MANIFEST.md: "This snapshot is what a paper should cite. \results/` at the repo root keeps changing as"`).

The gap: paper/results-frozen/MANIFEST.md (real, checked-in file) opens with exactly that sentence and has no disclaimer anywhere. This contradicts CLAUDE.md (repo policy, checked in): "paper/results-frozen/ ... is no longer under any special protection ... not the evidence backing a live citation." README.md's own "Reproducibility & Results History" section already correctly says "It's kept as historical reference, not as evidence for a live publication." A whole sweep of commits fixed this exact stale framing in README/CONTRIBUTING/CONTRIBUTORS/ROADMAP/CHANGELOG/website/llms.txt/faircode/benchmark.py when the freeze was lifted - scripts/freeze_paper_results.py was never touched by any of them, so it still generates a MANIFEST.md that reads like the freeze is still in force.

Repro:

$ head -5 scripts/freeze_paper_results.py
"""Freeze the current results/ into paper/results-frozen/ for a paper citation.
...
$ head -5 paper/results-frozen/MANIFEST.md
# Frozen results provenance

This snapshot is what a paper should cite. `results/` at the repo root keeps changing as
...
$ grep -n "no longer under any special protection\|not the evidence" CLAUDE.md
19:`paper/results-frozen/` is kept in the repo, but it is no longer under any special protection - it's
21:year's real paper run, not the evidence backing a live citation.

tests/test_freeze_paper_results.py's own docstring confirms this exact text is untested (only mirror_for_mcp is in scope there).

Fix direction: reword the module docstring and the generated MANIFEST.md text to match CLAUDE.md/README's current framing - a comparison snapshot for next year's paper run, not evidence for a live citation.

Start with the module docstring and MANIFEST.md generation around line 110 in scripts/freeze_paper_results.py, then compare the wording with CLAUDE.md and README.md. Update the stale framing in the generator and checked-in paper/results-frozen/MANIFEST.md, and inspect tests/test_freeze_paper_results.py to confirm the affected text is not currently covered.