Update benchmark

[dilithjay]

• Update benchmark with latest Gemini and GPT models
• Set default LLM model to gemini-3.5-flash
• Update AUTO mode results with new default LLM (gemini-3.5-flash)
• Add Claude models to benchmark

[Copilot]

Pull request overview

Updates Lexoid’s published benchmark artifacts and runtime defaults to reflect newer LLM offerings (Gemini/GPT/Claude), while extending the benchmark harness to support resumable runs and optional per-parse process isolation.

Changes:

• Refresh benchmark datasets/docs (CSV, README, Sphinx docs) with newer model results and added Claude models.
• Switch the library default LLM to gemini-3.5-flash and adjust Gemini/Claude request behavior accordingly.
• Enhance tests/benchmark.py with progress caching, token-usage persistence, and optional subprocess-isolated parsing.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/results.csv	Updates benchmark output CSV with newer model runs/results.
tests/benchmark.py	Adds parse isolation mode, caching of per-file/per-config results, and token usage capture.
tests/api_cost_mapping.json	Extends pricing map for newer Gemini/GPT/Claude models.
README.md	Updates benchmark leaderboard table to match new results.
lexoid/core/utils.py	Changes default LLM to gemini-3.5-flash.
lexoid/core/parse_type/llm_parser.py	Tweaks Gemini thinking config and adjusts Claude request parameters.
lexoid/core/conversion_utils.py	Preserves actual image MIME type in generated data URLs.
lexoid/api.py	Hardens bbox detection when segments may be missing/empty.
examples/example_notebook.ipynb	Updates example usage to gemini-3.5-flash.
docs/benchmark.rst	Updates benchmark documentation table with refreshed rankings/results.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.