feat: expand model comparison axes and free Zen coverage

[atomantic]

Summary

Comparison now supports independent X/Y selections for benchmark score, cost, speed, response time, and input/output prices. Changing metrics resets zoom; labels, units, tooltips and point filtering follow the selection while the evidence table retains its units.

Ship sourced pricing for all eight free endpoints advertised by the live OpenCode Zen catalog on September 6. An additive migration preserves existing research. Unknown quality, speed and quotas remain unknown; zero prices can be plotted with linear scaling. Document primary sources and an implementation path for coding, creative writing, image analysis and tool-use comparisons, without importing unverified benchmark scores.

CI also exposed a timing-dependent iMessage sync test. Replace its fixed 20 ms sleep with a persistence-readiness promise, preserving the concurrent-call assertion.

Test plan

• 13 comparison UI tests passed, including changed metric coordinates, zoom reset, exact endpoint coverage and zero-price plotting.
• 25 API/migration/catalog-scope/sync tests passed, including preservation of installed research and refusal to overwrite future schemas.
• Client production build and scoped lint passed.
• Optional OpenCode reviewer: inconclusive. Installed CLI help has no verified invocation-local read-only tool isolation; the required procedure forbids supplying review content without that boundary. No external review verdict is claimed.