Cut pagination off on createdAt, the field the ordering guarantees

[Thayorns]

total_releases_published asks for releases ordered by CREATED_AT DESC but decides when to stop paginating by looking at publishedAt. Those two fields part company on any release that was drafted long ago and published later: its publishedAt sits inside the window while its createdAt is old, or the other way round. When a whole page reads as published before since, the loop stops there and never reaches the pages behind it, undercounting the result.

The cutoff now reads createdAt, the field the ordering actually guarantees, and the query asks for it alongside publishedAt. Counting still uses publishedAt, which is the right field for the question being answered; only the decision about when to stop walking moved.

Three tests cover it. One puts an old-published, recently-created release on the first page and a countable release on the second, and fails on the old code because pagination stopped early. One walks the opposite shape, a recently-published but old-created page, and checks the scan now stops on it. One puts a release created exactly at since on the boundary and checks the scan keeps going.

Closes #661

[Thayorns]

@yegor256 fixed here, take a look please

[yegor256]

This touches the same total_releases_published boundary condition as #678, via a different approach — only one should be merged. Assigning to @yegor256 to reconcile.

---

Generated by Claude Code

[Thayorns]

@yegor256 take a look please