Document GPU allowlist requirement (refs #4)

[spacexbt]

Summary

• Adds a Supported GPUs subsection to README.md under Hardware Requirements, explaining that the auth service maintains an allowlist of validated GPU models on top of the 24 GB VRAM minimum, and how to request additions.
• Adds a matching admonition to docs/quick-start/requirements.md so the docs site is consistent.
• Adds a Troubleshooting entry for the exact server error string GPU device not in allowlist: '<model>', cross-linking back to the new subsection.

Motivation

Issue #4 reports an NVIDIA L4 being rejected even though it meets the documented 24 GB VRAM minimum, with no obvious next step from the docs. The actual allowlist lives in the authorization service (not in this repo), so a code-level fix isn't possible from here — but documenting that the rule exists and providing a clear path to request additions makes the situation legible to future users without committing the team to any specific allowlist policy.

Scope

Docs-only, +16 lines across 2 files, no behavior changes. The wording deliberately does not enumerate which specific GPU models are allowed — that's the team's call to publish if/when desired.

Test plan

• Render README locally — new section appears under Hardware Requirements with consistent heading depth
• Verify Troubleshooting entry matches the exact server error string from issue NVIDIA L4 (g6 EC2 instances) not in GPU allowlist #4
• docs/quick-start/requirements.md admonition matches the style of surrounding !!! info / !!! warning blocks
• Confirm mkdocs serve renders the new admonition correctly (no local Python 3.11 + mkdocs-material env on hand; happy to verify if a reviewer flags rendering issues)

🤖 Generated with Claude Code

[ryptotalent]

add a100 plz