feat: ez move restructures the GitHub native stack it moves within

[rohoswagger]

Closes the gap reported from a real stack: ez move --onto <parent> rebased locally and updated ez's metadata, then called update_pr_base, which GitHub rejected with "Cannot change the base branch because the pull request is part of a stack". The failure became a ui::warn, the command still printed Moved, and the receipt carried no remote status at all. Local and ez topology were right; GitHub's was stale, and nothing said so. Recovering meant gh stack unstack + gh stack link by hand.

What changed

Inserting a PR into the middle of an existing native stack is only supported by dissolving and recreating the stack — and ez already knows how, via repair_native_stack_exact behind ez sync --repair-native-stack. ez move now performs that restructure itself.

Detection before mutation. detect_native_stack looks the PR's stack up before anything local is touched, so the decision is made with full knowledge rather than discovered from a failed edit after the fact. A lookup error is non-fatal — ez move is primarily a local operation and shouldn't refuse to work because GitHub is unreachable.

No more doomed base edit. A PR in a native stack skips update_pr_base entirely; the restructure is what moves the base. That removes the misleading warning at its source.

Scoped to moves that involve a stack. The restructure runs only when the branch was actually in one. An early version reconciled every chain in the repo unconditionally, which made ez move fail on repos and GitHub installations with nothing to do with the move — caught by move_force_allows_descendant_merge_linearization.

Receipts split local from remote, which was the specific agent-facing complaint:

{"cmd":"move","branch":"feat/x","from":"main","onto":"feat/base",
 "local":"moved","restacked":1,
 "pr_base":"deferred_to_native_stack",
 "native_stack_before":{"number":88,"size":4}}

pr_base is one of updated, deferred_to_native_stack, {"status":"failed","detail":"..."}, or null when there's no PR. The restructure emits its own receipts under cmd: "move", and a restructure failure exits 2 after the local receipt — so an agent can always tell exactly how far the move got.

Refactor

Sync's native-stack reconciliation moved to cmd/native_stack::reconcile_stacks(state, cmd, retry_command, repair), so ez sync and ez move restructure identically instead of drifting apart the way merge's restack loop did.

Cost

One repo view round-trip per move on a branch with a PR — the probe that makes detection-before-mutation possible. Pinned in move_warns_when_pr_base_update_fails_but_persists_local_move.

493 unit tests + all 25 integration suites green, clippy clean.