make it work with lm-studio

i prefer lm-studio over chandra_vllm
because docker container is overkill

this should also work with other OpenAI-compatible backends

usage

1. start lm-studio
2. load the chandra-ocr-2 model
3. enable local server: developer → start server
4. run chandra

chandra test.png test.png.chandra \
  --method vllm \
  --vllm-api-base http://127.0.0.1:1234/v1 \
  --vllm-api-key lm-studio

error handling

a wrong vllm-api-base like http://127.0.0.1:1234
was giving the confusing error

raw = completion.choices[0].message.content
          ~~~~~~~~~~~~~~~~~~^^^
TypeError: 'NoneType' object is not subscriptable

so now this honors completion.error
and it shows a warning if vllm-api-base does not end with /v1

AVIF image format

fix #98