Give the mache pins a Dependabot pull request of their own

[aywrite]

The workflows pin mache's actions in thirteen places, and the last three moves of those pins were made by hand. Dependabot already watches the actions, but in one group with everything else, so a mache bump would arrive beside unrelated ones. mache's releases can change what a match plays: v0.6.0 moved the default book table, and the hand bump had to add six book_table lines to keep scripts/book.sh.

• .github/dependabot.yml: mache gets its own group and is excluded from the general actions group.
• docs/DEVELOPMENT.md: read mache's changelog for a breaking line before merging a bump.

Not checked: whether Dependabot matches aywrite/mache* against the path-style pins. The next mache release will show it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MStzRapxqqqByoFQmfA69p

---

Generated by Claude Code

[github-actions]

Speed against adfd9660

Measured on AMD EPYC 9V45 96-Core Processor.
Both sides built with rustc 1.98.1 (48a229cea 2026-09-01).

Both sides built and run on this runner in this job, the way
scripts/speed.sh measures a perf commit. Each round runs
both sides on a layout of its own, the same compiled code with
its code and data shuffled and moved, so the 95% interval carries
where the code landed as well as the run. The verdict holds the
interval against a 1% threshold. The default layout, the one a
release ships, is measured after as a diagnostic. The node
counts are the search's: they move when the search does, and a
speed change leaves them alone.

round     base nps  candidate nps  change
    1      7948505        7665596   -3.6%
    2      7885459        7841844   -0.6%
    3      7923549        7877993   -0.6%
    4      7926023        7907659   -0.2%
    5      7858360        7690200   -2.1%
    6      7838341        7862109   +0.3%
    7      7771421        7246211   -6.8%
    8      7528697        7828919   +4.0%
    9      7676436        7461271   -2.8%
   10      7738579        7525791   -2.7%
   11      7925959        7959193   +0.4%
   12      7884646        7880532   -0.1%
   13      7786138        7506493   -3.6%
   14      7536533        7273847   -3.5%
   15      7483743        7091441   -5.2%
   16      7682407        7221896   -6.0%
   17      7707567        7593010   -1.5%
   18      8012480        7953040   -0.7%
   19      7552122        7744908   +2.6%
   20      7406472        7383922   -0.3%
   21      7792515        7448146   -4.4%
   22      7553623        7383721   -2.2%
   23      7728975        7235016   -6.4%
   24      6970539        7104251   +1.9%
   25      7643381        7204080   -5.7%
   26      6497501        7456506  +14.8%
   27      6381422        7577512  +18.7%
   28      7726593        7353206   -4.8%
   29      7799075        6822161  -12.5%
   30      5885111        6569130  +11.6%
   31      6544040        7158251   +9.4%
   32      7649448        7277671   -4.9%
   33      7480487        7265706   -2.9%
   34      7379191        7450025   +1.0%
   35      7388128        7531092   +1.9%
   36      7052927        7198152   +2.1%
   37      7259614        7020848   -3.3%
   38      6799333        6585569   -3.1%
   39      6663799        7048969   +5.8%
   40      6982017        7647242   +9.5%

             nodes    time  median nps  faster half
base       5965973  0.78 s     7646414      7813124
candidate  5965973  0.80 s     7449086      7698047
change                           -2.6%        -1.5%

paired change -1.2%, 95% interval -2.5% to +0.8%
diagnostic, on the default layout alone -1.7%, 95% interval -5.2% to +1.6%, 13 rounds

not resolved: the interval reaches past ±1.0% without clearing it, so
more rounds are needed to say either way

Speed: -1.2% (bench nps, 95% interval -2.5% to +0.8%, 40 interleaved rounds over shuffled layouts vs adfd9660)

The instructions each side's bench executed, counted under
cachegrind. The count repeats to within a few hundred
instructions, so a small change here is a real one, but it
prices instructions only: cache misses, mispredicted branches
and where the code lands are the speed's to show.

instructions against adfd9660, one cachegrind run a side, bench

            instructions    nodes  per node
base       9,367,189,351  5965973    1570.1
candidate  9,367,189,394  5965973    1570.1
change            +0.00%             +0.00%

within ±0.7%, as far as edits not made for speed have moved the count