Updater v0.8.3: Polkit detection enhance

[Leay15]

This pull request updates dependencies and refines how the application detects graphical and user session environments, especially for polkit (policy kit) support. The main focus is on improving session detection logic, clarifying function responsibilities, and adding targeted tests.

Dependency updates:

• Updated versions for several dependencies in Cargo.toml, including chrono, clap, notify-rust, reqwest, serde_json, semver, and tokio for bug fixes and compatibility.

Session detection improvements:

• Renamed has_graphical_session to has_interactive_graphical_session to clarify its purpose, and updated all call sites to use the new name.
• Introduced a new function has_user_session_bus_for_polkit, which checks for a user session bus environment and is now used in polkit agent detection.

Testing:

• Added a unit test to verify that has_user_session_bus_for_polkit returns true when the appropriate environment variables are set, even without a display.

[avifenesh]

This is an auto review done by revuto.

---

One finding on the new test's env-var handling. The dependency bumps (Cargo.toml/Cargo.lock are consistent), the has_graphical_session → has_interactive_graphical_session rename (no stragglers), and the relaxed polkit gate (has_user_session_bus_for_polkit) all look correct.

[avifenesh]

This is an auto review done by revuto.

---

This test mutates process-wide env vars (DISPLAY, WAYLAND_DISPLAY, DBUS_SESSION_BUS_ADDRESS, XDG_RUNTIME_DIR) and restores them with manual save/restore code at the tail (lines 2809-2828). If the assert!(has_user_session_bus_for_polkit()) on line 2807 panics, the restore block never runs, leaking the removed DISPLAY/WAYLAND_DISPLAY and the faked DBUS_SESSION_BUS_ADDRESS/XDG_RUNTIME_DIR into other env-dependent tests in the crate (e.g. the feature_picker tests that read DISPLAY/WAYLAND_DISPLAY, and has_interactive_graphical_session callers). env_lock() only serializes access — it does not roll back leaked state — so a single failure here can cascade into misleading failures elsewhere.

Use a Drop-based RAII guard so restoration happens on every exit path including panic. The crate already has this exact pattern in codex_cli.rs (EnvRestoreGuard::capture(&[...])); reusing/lifting that (e.g. into test_util) would keep restoration panic-safe and drop the ~20 lines of manual restore here.

[ilysenko]

Thanks, the runtime change looks good and the updater tests pass locally.

One thing to fix before merge: the new test mutates process-wide env vars and restores them manually after the assertion. If the assertion panics, the restore block will not run and the modified env can leak into later tests. env_lock() only serializes access; it does not restore state.

Please use a Drop/RAII env restore guard here, or move the existing EnvRestoreGuard pattern from codex_cli.rs into test_util and reuse it.

[ilysenko]

Reviewed current head 6e43d6a. The env restoration issue is fixed with a shared RAII guard, GitHub checks are green, and local codex-update-manager tests pass.