Improve AI vision prompt for accurate chart recognition and data extraction

Enhanced the AI vision analysis to better recognize chart types, extract precise data values, and match visual styling:

Chart Type Recognition:
- Added comprehensive list of all supported chart types (pie, line, bar, column, area, scatter, bubble, geo, gauge, candlestick, timeline, combo, radar, polarArea)
- Improved COMBO chart detection with explicit instructions to identify mixed visualization types (columns + lines)
- Added chart type mapping for all Visualizer-supported types in parser

Data Accuracy Improvements:
- Instructed AI to interpolate values between gridlines instead of rounding
- Added guidance to study Y-axis scale and gridline intervals carefully
- Emphasized 5-10% accuracy requirement for data values
- Provided examples of correct interpolation (e.g., 60% between 10-20 = 16)

Visual Styling Detection:
- Enhanced legend position detection (right/left/top/bottom)
- Improved color extraction in exact order with hex codes
- Added chart-type-specific styling instructions (pie slice text, 3D detection, donut detection)
- Included styling examples for pie, bar/column, line/area, combo, bubble, geo, gauge, and scatter charts

Combo Chart Support:
- Added explicit combo chart detection rules
- Instructed AI to use "seriesType" and "series" object for mixed types
- Provided CSV and styling examples for combo charts
- Added combo chart-specific analysis instructions

Context and Safety:
- Added professional context to prevent OpenAI safety filter rejections
- Clarified this is for legitimate data extraction and visualization purposes
- Maintained concise prompt while preserving all essential instructions

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: vytisbulkevicius