Tolerate URL normalization differences in cache lookups

[crazytonyli]

Description

Two integration tests in wp_mobile_integration_tests/tests/test_remove_site.rs panicked with site should exist. The root cause was an interaction between two PRs that landed in sequence on trunk:

• PR Add new APIs for deleting site data from WpApiCache #1232 added test_remove_site.rs, which looks up the test site with the raw credentials string http://localhost, and a url_candidates workaround that tolerates trailing-slash differences but only on the delete path.
• PR Use SiteInfo to create WpService #1239 then changed WpService to accept URLs as Arc<ParsedUrl> and store them via ParsedUrl::url(), which normalizes http://localhost to http://localhost/.

Neither PR's CI saw the combination. Post-merge on trunk, the cache stored the normalized form (http://localhost/) while the tests still queried the raw form (http://localhost), and select_self_hosted_site_by_url's bare WHERE url = ? lookup returned None. The trailing-slash tolerance from url_candidates had never been wired into the read path.

Changes

Register a urls_eq(a, b) SQLite UDF that compares URLs via ParsedUrl parsing, replace WHERE url = ? with WHERE urls_eq(url, ?) in select_self_hosted_site_by_url, and remove url_candidates. Every URL-keyed read now uses a single tolerant mechanism, so historical caches stored in any normalization shape continue to match.

Test plan

CI green.

Changelog

• I've added an entry to CHANGELOG.md under ## [Unreleased], using the Keep a Changelog categories (Added, Changed, Deprecated, Removed, Fixed, Security). Prefix breaking changes with **BREAKING:**.

[oguzkocer]

I am reviewing this as-is, but I wonder if it'd be better to handle the existing data through a migration and normalize the insert/look-up instead of trying to work around it.

[crazytonyli]

My main concern is that we'll have new code to insert random strings as "site url". To ensure the stored url exactly matches ParsedUrl, we'd need to do something like this for each insertion/update:

// normalized the `url_str` input to another url string.
let url = ParsedUrl.parse(url_str) // TODO: error handling needed, even though 99.99% time we'll insert legal url strings.
let normalized = url.to_str()

sqlite.exec("INSERT ... VALUE (?)", noramlized)

I don't like this approach because:

1. We may need to surface the url parsing error to the function API. That introduce unnecessary noises, when the url input is almost certain a legit url.
2. It'd be easier to remember (or read/write) the new urls_eq function than the url parsing dance.

[wpmobilebot]

XCFramework Build

This PR's XCFramework is available for testing. Add to your Package.swift:

.package(url: "https://github.com/automattic/wordpress-rs", branch: "pr-build/1317")

_{Built from 84e0f98}