Get DeepSeek-OCR running on an NVIDIA Spark with Claude Code

Our rewrite with extra checks. The poster didn't test it.

Goal: Get DeepSeek-OCR running inside this Docker container on an NVIDIA ARM device and prove it works on a sample image.

Inputs (fill in or confirm):
- Repo and model links (github.com/deepseek-ai/DeepSeek-OCR, huggingface.co/deepseek-ai/DeepSeek-OCR)
- A test image path
- Working directory you may write to

Rules:
- First check the environment: GPU visibility, CUDA version, Python, disk space; report whether it's feasible before installing anything big.
- Use a virtual environment and pin versions that match the CUDA/ARM build.
- Keep notes in a markdown log of every command and fix as you go.
- Prefer official wheels; if you must build from source, say why.
- Log each install command and its result in a notes file.

Checkpoint: Ask before deleting files outside the working directory or upgrading system drivers.

If you get stuck: If a dependency has no ARM build, stop and list options rather than looping on installs. If a login, CAPTCHA or two-step code appears, stop and ask me to handle it; never guess or store passwords.

Verify before you finish: Run OCR on the sample image and show the extracted text.

Output: The setup log, a run script, and the OCR output.

Done when: a single command reproduces working OCR on the sample image.