afx cleanup deletes an unmerged builder branch on the orphan path, unlike the path #123 fixed

[pseudoseed]

Found by running the cleanup, not by reading it.

#123 made --force on an unmerged live-row builder remove the directory but keep the branch:

await removeOrphanWorktree(workspaceRoot, worktreePath, force, { deleteBranch: merged });

That is the property that makes --force safe to hand to someone in a hurry — 900 MB of directory goes, committed work stays recoverable.

The orphan path does not pass the option:

// cleanupOrphan
await removeOrphanWorktree(config.workspaceRoot, worktreePath, force);

and removeOrphanWorktree defaults it on:

const deleteBranch = options?.deleteBranch !== false;

So afx cleanup -p <id> --force against a row-less worktree deletes the local branch whether or not it is merged, which is the exact behaviour CMAP flagged on #123 and the builder fixed — on one of the two paths.

Observed

Cleaning 17 orphans here, 12 of which reported Branch is not merged, took the local builder/* branch count from 19 to 2.

Nothing was lost: all 17 branches exist on origin, and I verified the specific worktree HEADs are reachable there —

pir-4          aee79bb2d   reachable from origin/builder/pir-4
spir-52        53bb76946   reachable from origin/builder/spir-52
spir-83        583e8d715   reachable from origin/builder/spir-83
air-78         9e6f845ca   reachable from origin/builder/air-78
air-98         b2d506721   reachable from origin/builder/air-98
experiment-39  6fec5e348   reachable from origin/builder/experiment-39

— so this was recoverable because every branch had been pushed. A builder that never pushed would have lost its commits outright, and the message would still have said Worktree removed.

Fix

Pass the same option the live-row path passes:

const merged = await isWorktreeMerged(config.workspaceRoot, worktreePath);
await removeOrphanWorktree(config.workspaceRoot, worktreePath, force, { deleteBranch: merged });

removeOrphanWorktree already computes merged internally for its own refusal check, so the cleaner fix is probably to have it decide deleteBranch from that rather than take it as a parameter at all — one source of truth instead of two call sites that must remember.

Worth a test that asserts the branch survives an unmerged --force on the orphan path specifically. #123's test covers the live-row path and passes, which is why this got through.

[pseudoseed]

On it! Working on a fix now.