docs: rework README with quickstart, architecture diagram and task showcase

[wlgys8]

Closes #19

Summary

• Add light/dark architecture diagram (SVG + <picture> theme switching) after the intro: define once → train fast (SKRL / RSL-RL / FastSAC) → deploy anywhere, with built-in environment categories and a roadmap placeholder for future hardware
• Rework Quick Start: embed the microduck training console screenshot, TensorBoard curves, and the GitHub-hosted demo video (auto-embedded player in the replay section)
• Add two new sections after Quick Start following the user-guide style: Task Environments (quadruped / humanoid / WBT with gallery screenshots) and Built-in Robot Models (7 robots with renders, registry names, DoF)
• Key features: high-precision + high-performance MotrixSim wording with docs link; Sim-to-Real bullet now emphasizes one policy codebase via Sim2Sim (MuJoCo) and Sim2Real (real hardware)
• Fix stale package count (ten → nine) in AGENTS.md and CONTRIBUTING.md
• Track *.gif via LFS so the 5.5 MB hero GIF does not enter git history
• Add table of contents and center standalone figures; point gallery links to readthedocs latest

Notes

• docs/source/_static/videos/microduck-walk.mp4 intentionally left uncommitted (kept locally for upcoming user-guide pages)
• Recommend squash merge

[Copilot]

🟡 Changes recommended

The PR description states docs/source/_static/videos/microduck-walk.mp4 is intentionally uncommitted, but the PR adds it (as LFS), so the intent needs to be reconciled.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR refreshes the repository’s top-level documentation to provide a more visual “quickstart + showcase” entry point for MotrixLab, while also aligning the workspace/docs metadata and ensuring large media assets are tracked via Git LFS. It also tweaks scripts/view.py to avoid importing PyTorch eagerly and updates dependencies so the default training backend is available by default.

Changes:

• Reworked README.md and README.zh-CN.md with a new quickstart, architecture diagrams (light/dark via <picture>), and environment/robot showcases.
• Updated scripts/view.py to lazily import torch and provide a clearer error when torch-frontends are requested without PyTorch installed.
• Added/updated Git LFS tracking for media assets (GIF/PNG/MP4) and corrected “ten → nine” workspace package references in contributor docs.

File summaries

File	Description
uv.lock	Locks motrix-rl base dependency on torch with platform-specific sources (CPU vs cu128).
scripts/view.py	Makes torch import lazy and improves missing-torch error reporting when viewing torch-frontends.
README.md	New quickstart, media embeds, architecture diagram, and task/robot showcases (English).
README.zh-CN.md	New quickstart, media embeds, architecture diagram, and task/robot showcases (Simplified Chinese).
motrix_rl/pyproject.toml	Adds torch as the default training backend dependency and tightens formatting.
docs/source/_static/videos/microduck-walk.mp4	Adds the demo video as a Git LFS-tracked asset.
docs/source/_static/images/train-console.png	Adds the training dashboard screenshot (LFS pointer).
docs/source/_static/images/microduck-walk.gif	Adds the hero GIF (LFS pointer).
docs/source/_static/images/microduck-training-curves.png	Adds the TensorBoard curves screenshot (LFS pointer).
docs/source/_static/images/architecture-light.svg	Adds the light theme architecture diagram SVG.
docs/source/_static/images/architecture-dark.svg	Adds the dark theme architecture diagram SVG.
CONTRIBUTING.md	Updates references to the workspace package count (nine).
AGENTS.md	Updates references to the workspace package count (nine).
.gitattributes	Tracks *.gif via Git LFS to keep large GIFs out of git history.

Review details

• Files reviewed: 11/14 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.