Support transfer experiments with KataGo-raw

[norabelrose]

Unfortunately due to the apparent discrepancies between analysis and (our version of) match, it's become necessary to check that our attacks "transfer" to KataGo-raw.

The raw-gtp.cfg file is necessary because KataGo-raw doesn't support config options with player indices in GTP mode. Also it forces you to specify explicitly that ponderingEnabled=false.

[codecov]

Codecov Report

Merging #60 (ff47145) into main (085b57b) will not change coverage.
The diff coverage is n/a.

@@            Coverage Diff            @@
##              main       #60   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            4         4           
  Lines           81        81           
=========================================
  Hits            81        81           

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[AdamGleave]

LGTM modulo adding comment about why we're copy-pasting from base.cfg (and what we're modifying)

[tomtseng]

I'm having trouble figuring out what commit 42892e for KataGo-raw is — navigating to https://github.com/HumanCompatibleAI/KataGo-custom/tree/42892ea19b4256803ba7a0ff18d1096a84d11fe6 gives me a 404

[tomtseng]

I'm having trouble figuring out what commit 42892e for KataGo-raw is — navigating to https://github.com/HumanCompatibleAI/KataGo-custom/tree/42892ea19b4256803ba7a0ff18d1096a84d11fe6 gives me a 404

fixed, needed to pull the latest upstream changes from lightvector/KataGo into a branch in HumanCompatibleAI/KataGo-custom

I put it in branch master in KataGo-custom. The README says that this is supposed to track lightvector/KataGo, but we had one extra commit AlignmentResearch/KataGo-custom#2 in master. I'm guessing this was a mistake — the commit also got cherrypicked into stable by Tony. I removed that commit and just force pushed the latest commits from lightvector/KataGo