ci(ci): Offer the table size on the Strength workflow's dispatch tab

[aywrite]

hash_mb was an input of the Strength workflow's reusable form only, so a match started from the actions tab always played at 256 MB. A change to how the transposition table replaces entries only shows when the table is full, which at 10+0.1 means a smaller table, so the size is now a dispatch input with the same default of 256.

A dispatch input arrives as a string, and the batch workflow declares the size a number, so it is read back off the resolve job's outputs through fromJSON, as the elo bounds are. scripts/tests/test_strength_ladder.py caught the first version, which handed the raw input to the typed one.

It has already run: the table aging matches were dispatched from this commit, at 16 MB and at 256 MB, and every shard's manifest recorded the size asked for. The script tests pass (204; the five that need chess or numpy were not run locally).

🤖 Generated with Claude Code

https://claude.ai/code/session_01WmzY2g6SZGYszT5dnuQ6tw

---

Generated by Claude Code

[github-actions]

Speed against d9379f35

Measured on AMD EPYC 9V74 80-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      5697686        5748479   +0.9%
    2      5832308        5849572   +0.3%
    3      5710223        5704070   -0.1%
    4      5799661        5689481   -1.9%
    5      5781973        5734929   -0.8%
    6      5845354        5764537   -1.4%
    7      5643188        5725111   +1.5%
    8      5672408        5738973   +1.2%
    9      5762627        5682387   -1.4%
   10      5627495        5739398   +2.0%
   11      5762565        5782539   +0.3%
   12      5496257        5643193   +2.7%
   13      5712919        5839409   +2.2%
   14      5834053        5817657   -0.3%
   15      5770502        5679131   -1.6%
   16      5546120        5603857   +1.0%
   17      5438786        5492067   +1.0%
   18      5589503        5524683   -1.2%
   19      5484938        5716093   +4.2%
   20      5686471        5618116   -1.2%
   21      5682512        5597674   -1.5%
   22      5442194        5684326   +4.4%
   23      4894506        5096874   +4.1%
   24      5521487        5381727   -2.5%
   25      5644603        5626243   -0.3%
   26      5579978        5662335   +1.5%
   27      5673725        5768421   +1.7%
   28      5757910        5515867   -4.2%
   29      5709245        5700092   -0.2%
   30      5762098        5672953   -1.5%
   31      5582546        5469029   -2.0%
   32      5548580        5503431   -0.8%
   33      5565771        5431289   -2.4%
   34      5630385        5762120   +2.3%
   35      5496080        5640493   +2.6%
   36      5646617        5676354   +0.5%
   37      5659746        5680072   +0.4%
   38      5639496        5300213   -6.0%
   39      5655443        5668969   +0.2%
   40      5684298        5682853   -0.0%

             nodes    time  median nps  faster half
base       5965973  1.05 s     5657594      5739914
candidate  5965973  1.05 s     5679602      5740526
change                           +0.4%        +0.0%

paired change +0.1%, 95% interval -0.6% to +0.8%
diagnostic, on the default layout alone -1.2%, 95% interval -2.3% to -0.0%, 13 rounds

no change beyond ±1.0%: the whole interval is inside it

Speed: +0.1% (bench nps, 95% interval -0.6% to +0.8%, 40 interleaved rounds over shuffled layouts vs d9379f35)

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 d9379f35, one cachegrind run a side, bench

            instructions    nodes  per node
base       9,011,280,137  5965973    1510.4
candidate  9,011,280,166  5965973    1510.4
change            +0.00%             +0.00%

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