tweak: support comparison in ai charts

[djbarnwal]

Add support for comparison time range in some the AI generated charts

Part of https://linear.app/rilldata/issue/APP-725/ai-improve-analyst-agents-visualization-logic

Checklist:

• Covered by tests
• Ran it and it works as intended
• Reviewed the diff before requesting a review
• Checked for unhandled edge cases
• Linked the issues it closes
• Checked if the docs need to be updated. If so, create a separate Linear DOCS issue
• Intend to cherry-pick into the release branch
• I'm proud of this work!

[ericpgreen2]

Clean, well-scoped PR. The comparison plumbing leverages the existing CartesianChartProvider infrastructure (dual measures with __prev suffix, Vega transforms) — no new rendering code needed. The prompt examples and validation logic are thorough.

Schema nit — add required to comparison_time_range object

charts_json_schema.go — The comparison_time_range object doesn't declare "required": ["start", "end"]. The Go handler validates this server-side so it won't cause a runtime bug, but having the schema match the actual contract helps the LLM produce correct output. Currently the schema would accept {"comparison_time_range": {}} as valid.

"comparison_time_range": {
  "type": "object",
  "properties": { ... },
  "required": ["start", "end"],
  "description": "..."
}

Follow-ups (non-blocking):

• Explore link drops comparison — ChartContainer.svelte:120 hardcodes showTimeComparison: false. Now that AI charts can have comparison, clicking "Open in Explore" loses it. Pre-existing but newly relevant.
• area_chart exclusion — Area uses CartesianChartProvider like line/bar, so comparison likely works. If the exclusion is deliberate (overlapping fills?), a brief comment in the prompt explaining why would help the LLM avoid suggesting workarounds.
• Evals — The create_chart prompt is getting rich enough that eval coverage would be valuable. Tests asserting the LLM produces valid specs for key patterns (comparison bar chart, invalid chart type + comparison rejected, etc.) would catch prompt regressions early.