fix: remove stuck node upgrade-scheduled status

[jacderida]

The daemon flipped a running node to UpgradeScheduled from a 60s poll the moment its on-disk
binary version drifted, but only returned it to Running on process exit + respawn. When the exit
never followed (false-positive drift, an adopted node still alive, or a fast/Windows upgrade the
poll missed), the node stayed UpgradeScheduled indefinitely.

Remove the status entirely rather than fix it: the window it represented was tiny and not useful.
Upgrades are now detected purely at process exit by on-disk binary version drift for exit codes 0
(Unix) and 100 (Windows RESTART_EXIT_CODE) — which also fixes the pre-existing misclassification
of Windows and fast upgrades that the exit-code-0-only path missed.

Summary

• Remove the NodeStatus::UpgradeScheduled status variant, the spawn_upgrade_monitor 60s poll,
  mark_upgrade_scheduled, the pending_version field/plumbing (runtime state, NodeInfo,
  NodeStatusSummary, REST handlers, CLI), and the NodeEvent::UpgradeScheduled SSE event.
• Detect an upgrade at process exit by binary version drift for exit codes {0, RESTART_EXIT_CODE}
  and respawn directly via respawn_upgraded_node.
• Retain version refresh (respawn_upgraded_node) and the NodeUpgraded completion event.
• Backwards compatible: the status was never persisted (the registry stores only NodeConfig), and
  a #[serde(alias = "upgrade_scheduled")] on Running maps any value emitted by an older daemon
  during a rolling upgrade back to Running.
• ant-ui needs no changes (it is a thin proxy over the daemon API).

Test plan

• Manually tested auto-upgrade node transitions on Linux, macOS and Windows — upgraded nodes
  transition cleanly with the UpgradeScheduled status removed.
• cargo check --all-targets and cargo clippy --all-targets --all-features -- -D warnings pass.
• cargo test --all (full suite via CI).

🤖 Generated with Claude Code

[dirvine]

Approved under the requested severe-issue-only threshold.

Reviewed PR #145 at head a40b61a and did not find a PR-introduced severe blocker in the upgrade-status removal / exit-driven upgrade respawn path.

Verified locally:

• cargo check --all-targets passed
• cargo clippy --all-targets --all-features -- -D warnings passed
• cargo test -p ant-core upgrade --lib --no-fail-fast passed
• cargo test -p ant-core node_status --lib --no-fail-fast passed

Notes, not blockers for this PR:

• PR-head Security Audit is failing because the branch is based before the quinn-proto 0.11.15 lockfile fix on main; GitHub's synthetic merge ref has quinn-proto 0.11.15, so this does not reintroduce that advisory in the merge result.
• Follow-up issues should cover hardening upgrade-respawn against concurrent stop/eviction and updating docs/UI notes for the removal of UpgradeScheduled / pending_version.

Given Chris's stated release urgency and severe-only review bar, this is approved.