docs: add a supported models and hardware page

[twu3202]

Purpose

The second part of #45, as split there with MinhaoLi0318, who takes the quick start and installation pages.

• docs/supported-models.md: the compatibility table [Doc]: Community help wanted — documentation site and GitHub Pages #45 asks for. Each model and worker on main against CPU, NVIDIA CUDA and Apple Metal, with its requirements and a status (Validated, Unverified, Planned, Not supported) that links the recipe, the merged PR or the tracking issue behind it. Models still being added are linked through the new model issue label.
• mkdocs.yml: nav entries for the page and for the two Cua-S1 recipes, which were built but not in the nav.
• README.md: one line linking the page, in place of "CUDA and Metal coverage will be documented per model". The table rows are left to the model PRs.
• src/models/cua_s1/README.md: on Add Cua-S1 0.2 multimodal CUDA worker with upstream parity #12 you asked to keep the Python multimodal worker as a correctness reference and said "We should update the existing Cua-S1 plan to reflect this direction as well." The status still called the multimodal adapter deferred; it now points to that reference worker and says native execution of the adapter is not covered yet.

MinhaoLi0318's RTX 5090 and M5 Pro checks can update the status cells once they land with his pages.

Test Plan

System1-Omni Version / Commit: 2e33ef7, on main ce38770.

• mkdocs build --strict with docs/requirements.txt.
• Test merges with all open PRs (git merge-tree), compared with the same merges into main.
• The start and request commands of each recipe at ce38770, with ports 8831 and 8832, on one machine: AMD Ryzen Threadripper PRO 7965WX, RTX 6000 Ada (sm_89), Ubuntu 22.04, driver 595.91.07, CUDA 13.2, rustc 1.97.0. The Laya worker was installed as its recipe says. The two Cua-S1 workers used an existing environment that matches requirements-text.txt and the pinned weights, including a merged export from export_text_merged.py.

Test Result

• Strict build: no warnings. The two Cua-S1 recipes were outside the nav before; now only docs/benchmarks/cua-s1-cuda-graphs/README.md is.
• No open PR conflicts with this branch beyond its existing conflicts with main.
• Laya text worker on CPU (Python 3.12.13, laya 0.3.20, torch 2.14.1): compare_with_backend.py passed all five checks.
• Cua-S1 text reference worker: the recipe's request returned the recipe's example response byte for byte.
• Cua-S1 native text worker: the same request gave the same choice, with probabilities within 6e-5 of the reference.
• These runs back the Validated cells for LAYA on CPU and both Cua-S1 text workers on CUDA; the cells link the merged PRs that first recorded them.

Self-review

Before marking this PR ready for review or requesting maintainer review, complete
the self-review checklist.
Keep the PR in draft while this work is incomplete.
For agent assistance, use the optional precheck-pr skill.

• I have reviewed the full diff and addressed the issues I found.
• I have checked that the change follows the project's architecture and stays focused on the stated purpose.
• I have run the checks appropriate to this change and reported commands, results, and anything I could not verify above.
• I have checked that the PR description, documentation, and any accuracy or performance claims match the implementation and available evidence.