Links search matches the URL, and multi-word queries

1 · Search now looks inside the URL

The search box matched title, description, note, domain, site, author, handle and tags — but never the URL itself. A link you only remember by a piece of its address was unfindable. The raw URL is now part of what gets matched, so /reel/, an owner/repo pair, or a ?v=… id all work.

2 · Multi-word queries match term by term

Typing two words used to search for that exact phrase. Now each word is matched separately and all of them must appear somewhere on the link, in any order — across different fields. This only ever finds more than before, never less.

Also in this change

A titleless link no longer matches "undefined". The old code concatenated missing fields into the searchable text, so legacy links without a title carried the literal word. They don't any more.

The searchable text is built once per link instead of being re-assembled for every link on every keystroke.

No copy changed — nothing to publish to the Translations CMS, and no new env vars or secrets.

QA — try these yourself

New behavior

1. Open any Links Manager and search a path fragment (/reel/, /watch, a slug word) → only links whose URL contains it remain.
2. Search an owner/repo pair from a GitHub link → that link is found.
3. Type two words that live in different places (a domain word + a title word) → the link still matches.

Regression watch

4. still works: searching by title, note, tag, author, or site name.
5. still works: search combined with the Type / Source / Tags / Watched filters, and the result count line.
6. still works: pagination resets to page 1 when the query changes, and the "By author" view.
7. still works: Hebrew/RTL — the search box and results mirror correctly (no copy changed).

Deployed?