introduce modelpack for model in OCI format

[CormickKneey]

Summary

This PR enhances the ORAS Python client to support multiple compression formats by implementing suffix-based compression detection, replacing hardcoded media type comparison with a flexible approach. So we can use this to support various media types (for example, from https://github.com/modelpack/model-spec/blob/main/specs-go/v1/mediatype.go) instead of the original specific one.

Key Changes

1. New Compression Functions (oras/utils/fileio.py)

• extract_tar(): Handles uncompressed tar archives
• extract_tar_zstd(): Supports zstd-compressed archives (requires zstandard package)
• get_compression_from_media_type(): Detects compression format from media type suffixes
• extract_by_compression(): Generic extraction dispatcher

2. Enhanced Provider Logic (oras/provider.py)

• Replaced hardcoded media type check with suffix-based detection
• Dynamic compression format handling based on media type suffixes
• Proper file extension mapping (.tar.gz, .tar.zst, .tar)

Compression Detection

Uses suffix matching to identify formats:

• +gzip suffix → gzip compression
• +zstd suffix → zstd compression
• .tar suffix → uncompressed tar
• .raw suffix → raw files (no extraction)
• Default → gzip (backward compatibility)

Dependencies

• Optional: pip install zstandard for zstd support

[vsoch]

Is this something you've discussed with the primary oras maintainers in slack, or is this feature already implemented in oras go?

[tarilabs]

Suggested change

	annotation_filepath = "org.cnai.model.filepath"
	annotation_filepath = "org.cncf.model.filepath"

to align with latest since Sandboxing.

Ref: https://github.com/modelpack/model-spec/blob/d96289821ac99a777baaec74c8f3ee2625255adc/docs/annotations.md?plain=1#L9

[CormickKneey]

Is this something you've discussed with the primary oras maintainers in slack, or is this feature already implemented in oras go?

Sorry for the late response — I missed the email. I haven’t discussed this with the primary ORAS maintainers yet. Let me start a thread on Slack later this week!