Fixes #287

Changes proposed in this pull request:

• Forward unknown command line arguments to vLLM.

This might be a bit of a silly approach, but theres a ton of arguments for vLLM, and patching in argument by argument as a new one is needed seems unnecessary, so I'm just forwarding any unknown argument to vLLM. Happy to hear alternatives.

I was testing OlmOCR + vLLM 0.10.0 with data parallelism on some B200s as 0.10.0 has improved support for Blackwell, but I was running into issues with the multimodal input cache so I needed to add --disable-mm-preprocessor-cache (does solve the issue), this allows me to do it from the command line.

Edit: I also needed to increase the wait time for vLLM startup, but I didn't add that here. Reasonable to add as a command line arg?

Before submitting

• I've read and followed all steps in the Making a pull request
  section of the CONTRIBUTING docs.
• I've updated or added any relevant docstrings following the syntax described in the
  Writing docstrings section of the CONTRIBUTING docs.
• If this PR fixes a bug, I've added a test that will fail without my fix.
• If this PR adds a new feature, I've added tests that sufficiently cover my new functionality.