`nanoc` recursively executes itself to make another recursive call to bundler

The mean time for docs build over the last 10 runs is 16.87 minutes.

✗ curl -sSL https://gitlab.com/gitlab-org/gitlab-docs/-/pipelines.json | jq ".pipelines[].details.duration"
1109
935
1040
949
1115
868
1063
969
1076
1039
916
925
1064
1039
1074 

From which processing of .js with rollup takes more that 2 minutes (154.21s).

I tried to see if I can replace it with esbuild and spent few days researching how the rollup here is invoked. In pipeline config there is no separate minification step. At first I though that maybe I am looking at the wrong pipeline, because it looked like bundle nanoc should invoke bundle "script" from package.json, which redefines bundle command (which looks like a hack to me). I've just proved that I was right, but it required quite a lot of time to discover the nanoc frontend custom command that does this. I hoped to just swap nanoc to esbuild and see at least 1 minute of JS minification gone, but it is not that simple. I documenting it here, to improve the contributors experience (@nick_vh) by simplifying the pipeline.

Right now GitLab pipelines invokes bundle, which invokes nanoc, which invokes nanoc frontend, which invokes command/frontend.rb, which invokes yarn bundle, which invokes package.json:bundle script, which finally invokes rollup.

graph LR
  bundle --> nanoc --> n[nanoc frontend] --> y[yarn bundle] --> p[package.json:bundle] --> rollup

The proposal is make GitLab pipelines invoke bundle, which will invoke nanoc. Then execute rollup as a separate command from the pipeline.

graph LR
bundle --> nanoc
rollup

Will that break anything?