Use dedicated viewer bucket for datapackage and duckdb queries

[krivard]

Overview

Closes #78

What did you change in this PR?

• Fetch datapackage from viewer bucket (__init__.py)
• Run preview queries against viewer bucket (index.ts)
• If the user wants the whole parquet for offline analysis, use the nightly bucket (search_results partial; unchanged in this PR)

Testing

How did you make sure this worked? How can a reviewer verify this?

• Ran locally
• Verified server log included {"event": "Getting datapackage from https://s3.us-west-2.amazonaws.com/pudl.catalyst.coop/eel-hole/pudl_parquet_datapackage.json"
• In a browser, turned on network monitoring dev tools
• Loaded http://127.0.0.1:8080/search and previewed a table
• Examined headers for parquet requests (the ones with 206 response status); verified URL started with https://s3.us-west-2.amazonaws.com/pudl.catalyst.coop/eel-hole/

To-do list

• Review the PR yourself and call out any questions or issues you have

[e-belfer]

This worked as expected for me! I was able to reproduce your testing workflow locally. I've updated the uv lockfile to resolve failed builds: https://github.com/catalyst-cooperative/eel-hole/actions/runs/19572591605

and to handle an npm package that was triggering a dependabot warning. After doing this I reran the docker build and all seemed to work as expected.