Add support to use volumes as adapter source for model inference

[varkat77]

FineTuning output fine tuned data as volume using pvc, however it can't be used as adapter source for inference models.
https://github.com/kaito-project/kaito/blob/main/examples/fine-tuning/kaito_workspace_tuning_phi_3_with_pvc_volume.yaml

Please add support to use volumes as adapter sources for inference

Puller container in inference workspace pod is using 3P quay.io/skopeo/stable image which is non mcr image.

[kaito-pr-agent]

💡 Auto-generated documentation-based answer:

KAITO supports fine-tuning models through its CLI using the kubectl kaito deploy command with the --tuning flag. The tuning process:

1. Uses QLoRA method (--tuning-method qlora)
2. Accepts input data via URLs (--input-urls)
3. Outputs tuned adapters as container images (--output-image)
4. Requires registry credentials (--output-image-push-secret)

Example tuning command:

kubectl kaito deploy --workspace-name workspace-tuning-phi-3 \
  --model phi-3-mini-128k-instruct \
  --tuning \
  --tuning-method qlora \
  --input-urls "https://huggingface.co/datasets/philschmid/dolly-15k-oai-style/resolve/main/data/train-00000-of-00001-54e3756291ca09c6.parquet?download=true" \
  --output-image "myregistry.azurecr.io/adapters/phi-3-tuned:v1" \
  --output-image-push-secret my-registry-secret

Relevant Sources:

• https://github.com/kaito-project/kaito/blob/main/docs/proposals/20250630-kaito-cli.md#tuning-deployment