docs: cover example solutions and check design for scorers - #205

Closes DOCS-1324

Problem

Ten more docs evals are scoped. The RLS eval took several review rounds, and most of the findings were scorer defects a contributor can catch before asking for review. CONTRIBUTING.md doesn't say how. Guidance comes from the review of #168.

Solution

• Adds Checking your scorer: one green solution that passes every check, a few bad ones carrying one deliberate flaw each, and the expected failures written down before scoring. Split into solution shapes, what to write, how to score, and reading the results.
• Names three solution shapes, since the 44 evals don't share one. A workspace overlay for local-stack evals, a remote/-shaped seed for tools evals, and a report for the 8 evals scored on what the agent said rather than what it built.
• Adds Designing checks: decide what a check does when the object is absent, and find a class of object by asking the system that owns it. Checks about an object the prompt names can name it.
• Notes that a flaw usually trips several checks, and that transcript checks can't be exercised by a solution you wrote yourself.
• Lists solutions/ in the eval folder contents, above the optional entries, and links the new section from Submitting evals for review.

Manual testing

1. Open CONTRIBUTING.md on this branch. Designing checks and Checking your scorer sit between Writing scorers and Adding an experiment.
2. Click either Checking your scorer link in that page. Both jump to the new section.
3. Read the diff of the new content. See that it is technically accurate and helpful for all Eval contexts.