adopt-project.py check: refuse the branch names execute fails on after both legs exist (case, hideRefs, git's own rules) #520

A4: #382's plan-split-branch-taken compares branch names case-sensitively, so on macOS and Windows a source branch that differs from the split branch only by case passes check, and execute makes both legs before git refuses. Ruled 2026-10-12: compare case-folded on every OS.
— #457, body

adopt-project.py reads the branches of a local source or assembly with git ls-remote (_branches_at, adopt-project.py:3830, called at :3974, :4028 and :4034 by #382's _places_check_asks and _refuse_a_branch_in_the_split_branchs_way), so a branch the repository hides with uploadpack.hideRefs or transfer.hideRefs is not seen: check says plan ok, and execute --local-remote-dir pushes both legs before the split push fails on git's ref lock. Direction: read a local path's branches with git for-each-ref. From the review of #382 (codex at 5a1ffd5, confirmed by an independent verification; here only the git side was re-checked: git ls-remote prints nothing for a transfer.hideRefs branch that git for-each-ref lists, and execute was not run, there being no git-filter-repo in this container).
— #457, body

L2-1: A residual of #420. Once check asks git about a plan's tracking_branch: and adopt_branch:, execute still also holds both to checked_value's alphabet (scripts/repo_shape.py:94), and git accepts names outside it (a=b, a,b). So check keeps saying plan ok for a plan execute refuses before any leg; plan --tracking-branch a=b already refuses, so only a hand-edited plan reaches it. check should ask checked_value too, as it does for org:. Repro: set tracking_branch: "a=b" in a plan; check --plan exits 0 plan ok, execute --plan --yes --local-remote-dir D exits 2 REFUSED unsafe-value. Seen at main 580d7df.
— #457, comment