ci(release): count osx-arm64 in the Anaconda completeness check (#1922)

[JarryShaw]

• Searched for similar pull requests

• Followed the coding style (make pylint, make mypy, make isort)

• make test passes, and a test case covers the change

• Added a changelog entry — N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

• ci — workflows or build tooling

version_check's Anaconda completeness check only counted linux-64 and osx-64. A release whose osx-arm64 upload had failed still read PCAPKIT_CONDA_COMPLETE=true, so the conda gate skipped it. Per the ruling on #1922, osx-arm64 now counts too.

A single PCAPKIT_CONDA_SUBDIRS list now sets both the expected count and which platforms are counted. A new test pins that list to the conda job's os matrix and its target_platform mapping. It also runs the step against a listing with osx-arm64 missing, which fails on main. releasing.rst is updated to match. 1.5.0b13 already has all three platforms on Anaconda.

Closes #1922

[JarryShaw]

GOOD TO GO at 00799a72a. A Sonnet cross-review checked the conda matrix: it has no exclude and one Python list for every leg, and the case arms map to exactly linux-64, osx-64 and osx-arm64. So subdirs times builds is the right count. The exported list reaches the python3 - pipe under set -e, and the gate and stranded-release check only read the boolean. Restoring main's workflow fails three of the new tests, and so does dropping a matrix leg. The PR merges cleanly with #1926. One thing to expect: a release published earlier with only two subdirs will now read incomplete, and the next run uploads just the missing osx-arm64 leg.