[codex] Add GPU sandbox client support

[thedebugger]

Summary

Adds GPU sandbox creation support across the Tensorlake clients:

• CLI: tl sbx create/new --gpus <count> --gpu-model <model>
• Rust cloud SDK request model: serializes resources.gpus as an array of { count, model }
• Python SDK: gpus= and gpu_model= on sync, async, and Sandbox.create
• TypeScript SDK: gpus and gpuModel

The CLI/SDK surface exposes a single GPU model request and sends it as a one-element resources.gpus array, matching the server request shape.

Validation

Local checks:

• cargo test -p tensorlake-cli gpu -- --nocapture
• cargo check -p tensorlake -p tensorlake-rust-cloud-sdk-py -p tensorlake-cli
• make build_rust_py_client
• PYTHONPYCACHEPREFIX=/tmp/tensorlake-gpu-pycache poetry run python -m unittest tests.sandbox.test_client_rust_backend
• PYTHONPYCACHEPREFIX=/tmp/tensorlake-gpu-pycache poetry run python -m unittest tests.sandbox.test_client_rust_backend_async
• npm -C typescript test -- client.test.ts
• npm -C typescript run typecheck
• npm -C typescript run build
• git diff --check

Live dev smoke:

• CLI created GPU sandbox, ran nvidia-smi -L, then terminated it
• Python SDK created GPU sandbox, ran nvidia-smi -L, then terminated it
• TypeScript SDK created GPU sandbox, ran nvidia-smi -L, then terminated it

All three live sandboxes reported GPU 0: NVIDIA A10.

[eabatalov]

Would probably be helpful to have a list of exactly supported GPU models so users don't have to guess what to put here. Also presence of gpu_model can enabled --gpus 1 by default.