docs: add evaluation reproduction notes

[Rohithmatham12]

Summary

• add a top-level evaluation/reproduction FAQ covering requirements, prompt templates, deterministic inference settings, history formatting, coordinate handling, action mapping, mobile scroll vs drag, and deployment endpoint checks
• link the new guide from the main README header
• document the public action-space mappings previously clarified in issue discussions
• clarify that browser benchmarks should be run as multimodal GUI tasks and that exact reproduction depends on matching preprocessing, prompt, parser, history window, endpoint, and benchmark state

Issues

Addresses #246.
Addresses #247.
Addresses #221.
Addresses #169.
Addresses #96.
Addresses #86.
Helps #225.
Helps #121.
Helps #143.
Helps #122.
Helps #127.
Helps #150.
Helps #215.

Testing

git diff --check
python3 doc sanity check for README_evaluation.md and README link

This is docs-only; I did not run model benchmark jobs locally.

[CLAassistant]

All committers have signed the CLA.