This PR implements docker-style progress bars and resumable download functionality for the llama-server -dr flag, which downloads models from Docker Hub using OCI protocol.

Problem

Previously, when using llama-server -dr to download models from Docker Hub:

• No progress indication - users had no visibility into download status
• No resume capability - interrupted downloads had to restart from scratch, wasting bandwidth and time
• No integrity verification on cached files

Solution

1. Docker-Style Progress Bars

Added real-time progress display showing:

• Layer digest (short form: first 12 chars)
• Visual progress bar (50 characters wide)
• Percentage complete
• Downloaded size / Total size in MB
• Current download speed in MB/s

Progress updates every 100ms for smooth visual feedback and automatically detects TTY to only show on interactive terminals.

Example:

1234567890ab: Downloading [===================>              ] 39.0% (39.00 MB / 100.00 MB) 20.10 MB/s

2. Resumable Downloads

Implemented automatic resume capability:

• Partial downloads saved with .tmp extension
• Layer digest stored in .digest file for integrity verification
• Automatically resumes interrupted downloads from last position
• Efficiently skips already-downloaded bytes using io.CopyN
• Verifies file size matches expected size before finalizing
• Handles digest changes (model updates) by starting fresh download

Example resume:

1234567890ab: Resuming download from 39.00 MB
1234567890ab: Downloading [===================>              ] 41.0% (41.00 MB / 100.00 MB) 18.50 MB/s

3. Enhanced Cache Validation

• Validates cached models using stored digest
• Detects and handles digest mismatches
• Cleans up invalid partial downloads
• Uses atomic file operations (rename from .tmp to final)

Technical Details

Implementation

• Added progressWriter struct in Go that wraps io.Writer
• Uses sync/atomic for thread-safe progress tracking
• Progress written to stderr to not interfere with program output
• Progress updates throttled to 100ms intervals to avoid excessive I/O

Files Modified

• oci-go/oci.go (+180 lines): Core implementation
• docs/oci-registry.md (+43 lines): Documentation updates

Code Quality

• ✅ Go fmt applied
• ✅ Go vet passed
• ✅ Code review passed - no issues
• ✅ CodeQL security scan passed - no vulnerabilities
• ✅ Builds successfully on Linux x86_64

Benefits

1. Better User Experience: Users can see download progress and estimate completion time
2. Saves Bandwidth: Resume capability prevents re-downloading large files
3. Saves Time: Interrupted downloads can be resumed instantly
4. Better Reliability: Digest verification ensures file integrity
5. Familiar Interface: Progress format similar to Docker pulls

Backward Compatibility

Fully backward compatible:

• No breaking changes to existing functionality
• Works with existing cache files
• Handles both new and old partial downloads gracefully
• No changes to command-line interface

Usage

No changes required - existing -dr flag now includes progress and resume:

# Download with progress bar (interruptible and resumable)
./llama-server -dr ai/smollm2:135M-Q4_0

# Works with all existing -dr options
./llama-cli --docker-repo myuser/model:tag

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