feat: add v1.3-aligned MCP and pdf2vqa workflows

[LeoMY1]

Summary

This PR carries forward the work from dptech-corp#22 and aligns it with the UniParser v1.3 API and data model introduced by dptech-corp#23.

It adds two user-facing workflows: a simplified UniParser MCP server and a pdf2vqa extraction pipeline.

Changes

MCP server

• Refactor the MCP server to expose a single uniparser_parse tool.
• Support local PDFs, local images, and public PDF URLs.
• Align task submission, TOS upload, polling, and result retrieval with UniParser v1.3.
• Persist Markdown and pages_tree.json results locally.
• Return structured output paths and a content preview.
• Remove the legacy config.yaml-based configuration.
• Update Cursor and Claude Code setup documentation.

pdf2vqa

• Add the uniparser-agent package with parse and vqa CLI workflows.
• Extract questions, short answers, solutions, and related images from exercise and answer PDFs.
• Support:
  • Single PDF or image input
  • Public PDF URLs
  • Separate question and answer PDFs
  • Reusing an existing pages_tree.json
• Generate JSONL, Markdown, ShareGPT, run metadata, and raw LLM response files.
• Preserve LaTeX expressions in extracted answers.
• Keep intermediate results for debugging and repeated extraction.

Output safety

• Remove the overwrite option from the CLI, MCP, and Agent interfaces.
• Never reuse, delete, or overwrite an existing output directory.
• Automatically create a suffixed sibling directory such as results_1 or results_2 when the preferred directory already exists.
• Reject root, HOME, the current working directory, and Git metadata directories as output targets.

Tests and documentation

• Add unit tests for MCP input validation, parsing options, output persistence, and response schemas.
• Add Agent tests covering v1.3 API compatibility, VQA extraction, image export, PDF merging, LLM configuration, and output safety.
• Extend CI to run the main package, MCP, and Agent test suites.
• Update the main README, MCP documentation, Agent documentation, and Skill references.