NEWS.md 0.16.0: benchmark bullet splices two runs' numbers mid-sentence #76

If we do it: the 0.16.0 release note reads as written, and anyone quoting the
dft_rast_break_category() benchmark gets one run's numbers rather than two spliced
together. If we never do: a published changelog carries a sentence that breaks
mid-clause, and the two timings in it cannot be told apart.

The benchmark bullet in NEWS.md under # drift 0.16.0 contains a duplicated fragment —
two runs' numbers spliced mid-sentence:

…peak RSS 8.79 GiB across the whole run — every figure from timings.csv and
rss_summary.csv in that directory**.5 s**, dft_rast_break_category() 26.5 s, a
reconciliation crosstab 16.3 s, and peak RSS 8.82 GiB…

So the bullet reports both 8.79 and 8.82 GiB, and both 26.8 s and 26.5 s, with the
sentence broken at directory.5 s.

Needs a decision rather than a mechanical fix: which run the committed evidence in
data-raw/logs/benchmark_break_category/ actually describes, and therefore which pair of
numbers is correct. Read timings.csv and rss_summary.csv there and keep the pair that
matches, then repair the sentence.

Noticed during #73; unrelated to it, so it was left alone rather than folded into that PR.