[feat]: wrap into a python package

[mlisovyi]

Summary

Instead of using docker to run, allow to pip/uv-install a package and run the fastapi app locally.

Use case

This would allow to try it out (or use for smaller use-cases) without a docker installed in the system

Proposed approach

No response

Design constraint check

• I have read CONTRIBUTING.md and this proposal does not conflict with the load-bearing design constraints.

[gordonmurray]

Thanks, yes, this is a direction I'd like to take. The codebase is in good shape for it: the app is plain FastAPI with no frontend build step, and it already detects lancedb capabilities at runtime rather than assuming a pinned
version.

Rough plan:

• restructure backend/ into a proper package, with web/vanilla bundled as package data
• console entry point, something like lance-data-viewer /path/to/data --port 8080
• version read from the VERSION file, published to PyPI on release tags

The main design question is the lancedb dependency. The Docker images come in seven variants because older Lance data needs older lancedb pins, and a single package can't replicate that. My current thinking is a loose requirement (roughly lancedb>=0.29), making the pip package the "modern formats" path and keeping the version-tagged Docker images as the answer for legacy data. Would that cover your use case, or were you hoping to run it against an older lancedb you already have installed?

There are a couple of fixes I want to land first, but this is on the list.