Table Structure Recognition (TSR) skill to extract structured tables from table images. Use when the user provides a table image and asks to convert it into OTSL and/or HTML, validate table structure, or troubleshoot TSR output. Supports two backends: (1) dedicated TSR VLM at vllm-vlm-tsr-1 (model agilesoda/vlm-tsr) and (2) perception VLM at vllm-vlm-perception-1 (model nanonets/Nanonets-OCR2-3B) which may return HTML directly or OTSL-like tags that need conversion.