Browser observation drops tables and images

[harneet2512]

Problem

When observing pages with complex layouts (dashboards, data tables), the accessibility tree extractor misses table data and image alt-text. This means the agent loses context from visual-heavy pages.

Current Behavior

• Tables: only captures header row, body rows dropped
• Images: alt-text not captured, falls back to [image]
• Charts: completely invisible to observation

Expected

• Tables: full row/column data with [source: table] label
• Images: alt-text + surrounding context with [source: image] label
• Charts: axis labels + legend text with [source: chart] label

Note: MemContext stores raw content with labels — extraction intelligence comes from the calling AI, not from us.

Priority: Medium