docs(claude): add Error Messages four-ingredient strategy

[John-David Dalton (jdalton)]

Summary

Adds an Error Messages section to CLAUDE.md with the four-ingredient strategy we're using across the fleet (copied from socket-repo-template):

1. What — the rule that was violated (not the symptom)
2. Where — the exact flag / file / key (never "somewhere in config")
3. Saw vs. wanted — the offending value and the allowed shape
4. Fix — one concrete action to resolve it

Plus rules on imperative voice, naming colliding records, and not silently auto-correcting.

Why

Right now socket-cli has a lot of errors like Invalid value for --limit: foo — true but not helpful. After this PR lands, the agent has a written rule to reach for so error messages are consistent and actionable.

What's NOT in this PR

Zero source-code changes. This is just the strategy doc. The actual error-message cleanup across src/ will land in follow-up PRs split by area (commands, dlx, utils, env, test, etc.) so each PR stays reviewable.

Test plan

• Read the new section and make sure the examples feel right for socket-cli (some use InputError, one uses AuthError, each shows the 4 ingredients).
• Nothing else to test — docs only, no code changes.

---

Note

Low Risk
Docs-only change to CLAUDE.md with no runtime or behavioral impact. Low risk aside from potentially influencing future error wording conventions.

Overview
Adds a new ### Error Messages section to CLAUDE.md that standardizes how CLI errors should be written using a four-part format (what/where/saw-vs-wanted/fix) and a few concrete rules (imperative voice, avoid vague “invalid”, name collisions explicitly, don’t silently auto-correct), with good/bad examples using InputError/AuthError.

^{Reviewed by Cursor Bugbot for commit 31b7466. Configure here.}