v4.6.1 — the batched translation, and the per-distinct allele strip

Everything in 4.6.0, which built every wheel and published none. See the bottom for why.

---

Why 4.6.0 shipped nothing

It carried a test_changelog_covers_every_release that asserted the checkout has release tags.
actions/checkout clones without them, so the assertion fired inside a fixture and errored all
four Python jobs on both platforms — and publish.yml runs that suite as a precondition for
upload, so every wheel built and Publish To PyPI was skipped.

The test is deleted, not repaired: it was added unprompted during a changelog cleanup and was
never part of the request, and it reached outside the working tree for information tests have no
business depending on — which is exactly why it passed locally and failed in CI. The
fetch-depth: 0 added to both workflows to feed it is reverted, so they are byte-identical to
before.

The changelog content fixes from that cleanup stay, because those were the request: the missing
3.9.2 and 3.17.0 sections, the stray second ## Unreleased folded into the 3.14.0 that shipped it,
3.0.0's date, and the NOT PUBLISHED marker on 4.4.0.

4.6.0 is not on PyPI and its GitHub Release is withdrawn; the tag stays for the history.

1197 tests pass; ruff and sphinx -W clean. Full detail in
CHANGELOG.md.