post-release 26.10.1: the CI summary counts xfails apart from skips, release.md gets what the rehearsal found, the PyPI verification script moves into the repo, known issue for the JPype 1.7.1 leak #215

Four small follow-ups of the 26.10.1 release, in one change so CI runs once.

The bindings test summary printed "14 skipped" for 14 strict xfails (JUnit counts an expected failure as skipped), which reads as tests that do not run. It now prints N passed, K expected failures (xfail), 0 skipped, 0 failed and has a row for the xfails; the skip gate already excluded them.

docs/development/release.md lacked three things the rehearsal and the release found: the check that the tag is not upstream's release commit, --cleanup=verbatim for git tag -F (it strips the ## headings otherwise), and the step that downloads the 19 CI wheels and checks their sizes before tagging.

scripts/after_tag_verify.sh (the end-to-end PyPI check the release used: publish job conclusion, file set, sha256 against the artifacts, pip download on four platforms, fresh installs, docs) moves into the repo and into release.md; it stays out of the upstream pull request.

docs/guide/known-issues.md gets an entry for the JPype 1.7.1 leak of a Python object per boxed number (3.9 per row on to_list(), Result.get(), to_dict(), iter_dicts(); fixed on JPype master; to_json_list() and to_columns() are not affected), and the Ctrl-C entry says JPype master is unchanged.