Implement resumable downloads using HTTP Range requests

[Copilot]

Downloads interrupted by network issues or user cancellation currently restart from scratch. When pulling large models (e.g., ai/smollm2:135M-Q4_0), interrupting and resuming results in downloading from 0 bytes again instead of continuing from the partial download.

Changes

ResumableTransport wrapper (pkg/distribution/internal/store/resume.go)

• Intercepts blob GET requests and adds Range: bytes=N- headers based on partial file size
• Detects when server returns 416 or 200 (doesn't support Range) and retries without Range header
• Tracks resume success via DidResume() to coordinate with file writing logic

Store layer modifications (pkg/distribution/internal/store/blobs.go)

• Detects .incomplete files before initiating downloads
• New writeBlobResumable() appends to partial files when resume succeeds, overwrites when server doesn't support Range
• writeLayer() coordinates resume offset between transport and file I/O

Distribution client wiring (pkg/distribution/distribution/client.go)

• Wraps base HTTP transport with ResumableTransport
• Injects wrapped transport into both registry client and local store

Fetcher compatibility (pkg/go-containerregistry/pkg/v1/remote/fetcher.go)

• Accepts both HTTP 200 and 206 status codes
• Properly handles Content-Length for partial content responses

Behavior

$ ./model-cli pull ai/smollm2:135M-Q4_0
384a89bd054c: Downloading [========>                  ]  16.22MB/91.73MB
^C  # Interrupt

$ ./model-cli pull ai/smollm2:135M-Q4_0
384a89bd054c: Downloading [========>                  ]  16.22MB/91.73MB  # Resumes from 16.22MB

Gracefully falls back to full download when registry doesn't support Range requests (returns 416 or ignores Range header).

Original prompt

Implement resumable downloads

When someone executes:

./model-cli pull ai/smollm2:135M-Q4_0

And that pull gets interrupted. If we pull again, it should resume where we left off.

It can be tested like so it currently doesn't work, when I start a pull, stop it with Ctrl-C and start it again, it doesn't resume, it
downloads from scratch again:

$ MODEL_RUNNER_HOST=http://localhost:13434 cmd/cli/model-cli pull ai/smollm2:135M-Q4_0
384a89bd054c: Downloading [========> ] 16.22MB/91.73MB
$ MODEL_RUNNER_HOST=http://localhost:13434 cmd/cli/model-cli pull ai/smollm2:135M-Q4_0
384a89bd054c: Downloading [=> ] 2.306MB/91.73MB

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.