fix(opds): support mixed feed items parsing

[zeedif]

This is a follow-up to PR #14 to refine the handling of mixed OPDS feeds.

What's changed

• Robust Feed Separation: Even though the OPDS 1.2 spec implicitly states that a feed should be strictly either Acquisition or Navigation, some servers (like Copyparty) send mixed feeds containing both books and folders in the same response. I've modified the getFeed parsing logic to intelligently segregate publications and navigation arrays on a per-item basis instead of relying entirely on the very first entry. This restores support for such non-standard yet practical implementations found in the wild.

[chrox]

Please test and make sure it works as expected at least for:

• Calibre / CWA
• Project Gutenberg
• Standard Ebooks
• ManyBooks

[zeedif]

@chrox I have tested the changes against the standard OPDS catalogs (Project Gutenberg, Standard Ebooks and ManyBooks), and the parser continues to work as expected for all of them.

The original code determined the type of the entire array based solely on the first item (items[0]?.metadata ? 'publications' : 'navigation'). For standard, homogeneous feeds, this PR behaves exactly the same as before, placing all items exclusively into either publications or navigation. However, for mixed feeds, it now evaluates each item individually to separate them correctly.

I suggested this change specifically because Readest already evaluates this mixed-feed scenario (like Copyparty) in its test suite, specifically in apps\readest-app\src\__tests__\utils\opds-feed.test.ts.

I leave the videos below demonstrating that Readest works exactly the same with these catalogs after the changes.

• Project Gutenberg

Grabacion.2026-04-27.111218.mp4

• Standard Ebooks

Grabacion.2026-04-27.111312.mp4

• ManyBooks

Grabacion.2026-04-27.111403.mp4

[chrox]

Looks great. Thanks.