Release @corbits/evals 0.2.0 #31

Goal

Ship @corbits/evals 0.2.0. The stack is breaking (runEvaluation, expectedOutput, judge workflow), so this is not a 0.1.3 patch.

Scope

• Cherry-pick the open feature stack (PRs feat!: run evals with one runEvaluation call #23-docs: describe the current eval API #30), including the judge, trajectory, CLI, and docs fixes that landed after later pulls forked.
• Do not include the 0.1.3 / @corbits/system-one ^0.3.2 bump from chore(deps): bump @corbits/system-one to 0.3.2 (CL-11510) #22.
• Set the package to 0.2.0 and depend on @corbits/system-one ^0.4.0.
• Do not merge the superseded feature PRs separately.
• Do not npm publish until @corbits/system-one 0.4.0 exists on npm.

Done when

• release/0.2.0 is 0.2.0 and depends on @corbits/system-one ^0.4.0.
• The lockfile is updated only after 0.4.0 resolves.
• The release PR is open and not merged until that publish.