Add 'describe' or 'text' render output format for AI agents

Problem

AI agents need a compact, human-readable text representation of dashboards for reasoning, comparison, and summarization. The JSON format (PR #623, merged) provides structured data but is verbose and not designed for direct LLM consumption. A text format that templates over the JSON output would give agents a token-efficient markdown summary without any LLM dependency.

Context

• Foundation: format="json" now exists (PR #623). It walks the layout tree, executes queries, resolves charts, and returns resolved objects + data as JSON. The text format builds on top of this.
• dataface/core/render/json_format.py — render_face_json() returns the structured data we'll template over
• dataface/core/render/renderer.py — format dispatch (add elif format == "text")
• dataface/core/render/terminal.py / terminal_charts.py — existing terminal text rendering (different purpose but similar pattern for wiring)
• dataface/ai/mcp/tools.py — render_dashboard(format=...) already supports format param
• dataface/ai/tool_schemas.py — RENDER_DASHBOARD schema enum needs "text" added
• apps/playground/routes.py — playground render dispatch needs "text" case
• apps/playground/static/index.html — format-selector dropdown needs "Text" option
• dataface/cli/commands/render.py — CLI already handles json/terminal stdout path
• dataface/cli/main.py — --format help text needs "text" added
• No LLM or API key required — purely deterministic template formatting

Possible Solutions

1. Recommended — Deterministic markdown template over JSON output. Call render_face_json() internally (or reuse its tree walk), format the resolved data into compact markdown. For each chart: show type, field mappings, data summary (row count, value ranges for numerics, distinct values for categoricals). For KPIs: show metric name and value. No LLM, no API key, fast and free.

2. LLM-powered natural language insights ("Revenue grew 30% QoQ..."). Requires API key, adds latency and cost. Good as a future insights mode but wrong for this task.

Plan

Files to create: - dataface/core/render/text_format.py — render_face_text() that takes the JSON dict (from _face_to_dict) and produces markdown text - tests/core/test_text_format.py — tests for text output content

Files to modify: - dataface/core/render/renderer.py — add elif format == "text" branch - dataface/cli/commands/render.py — text prints to stdout like json/terminal - dataface/cli/main.py — update --format help text - dataface/ai/tool_schemas.py — add "text" to format enum - apps/playground/routes.py — add text format case in render dispatch - apps/playground/static/index.html — add Text option to format-selector dropdown

Example output:

# Revenue Dashboard

## Revenue Over Time (line)
- x: month, y: revenue
- 12 rows | month: Jan–Dec | revenue: $100K–$1.2M

## Total Revenue (kpi)
- metric: revenue = $4,200,000

## Sales by Region (bar)
- x: region, y: sales, color: category
- 8 rows | 4 regions | sales: $50K–$800K

Steps: 1. Write failing test: render a face with format="text", assert output contains chart type, field names, data summary 2. Implement render_face_text() — reuse _face_to_dict from json_format, template to markdown 3. Wire into renderer.py, CLI, MCP schema, playground routes + HTML dropdown 4. Run just ci

Implementation Progress

• GitHub issue: https://github.com/fivetran/dataface/issues/369
• Foundation: format="json" merged in PR #623
• Created dataface/core/render/text_format.py — reuses _face_to_dict from json_format.py, templates into compact markdown
• Wired format="text" into renderer.py (early return before SVG, like JSON)
• CLI: text prints to stdout (dataface/cli/commands/render.py), help updated in main.py
• MCP: added "text" to format enum in dataface/ai/tool_schemas.py
• Playground: added text case in apps/playground/routes.py (wrapped in <pre>) and <option> in static/index.html
• Tests: 10 tests in tests/core/test_text_format.py covering face title, chart types, field mappings, row counts, numeric ranges, categorical values (few/many), KPIs, nested faces, multiple charts
• All 2738 tests pass; formatting and mypy clean on all changed files

QA Exploration

• [x] QA exploration completed (N/A — rendering/CLI/tooling task validated by tests and playground route wiring)

Review Feedback

• [x] Review cleared