fix: guard ROS plugin updates during context shutdown

[yeseorizi]

Problem

Gazebo may still run an update callback after the shared ROS context has shut down. Calling spin_some() or publishing at that point can throw an RCLError. A plugin must also leave process-level signal handling to Gazebo instead of replacing it during rclcpp::init().

Scope

Guard five plugin implementations: SphericalCoords, OceanCurrentPlugin, UsblTransceiver, UsblTransponder and SubseaPressureSensorPlugin.

• Initialize ROS with SignalHandlerOptions::None.
• Check the node's context before ROS operations.
• Suppress an RCLError only when shutdown invalidated that context; rethrow it while the context is valid.
• Add a pressure-plugin regression for preserving the SIGINT handler and tolerating a post-shutdown update.

The USBL and pressure guards extend this existing PR rather than opening another overlapping shutdown PR. Latest main was merged normally into this branch. No launch timeout, Docker/CI or MAVROS changes are included. The depth timestamp fix is separate dependent PR #19.

Validation

Linux ARM64, Ubuntu 26.04, ROS 2 Lyrical / Gazebo Jetty:

• Built posim_gz_sensor_plugins and posim_ros_gz_plugins from this branch.
• Camera regression: 5/5 cases; new pressure shutdown regression: 1/1.
• Control run using the unchanged main pressure implementation failed the new shutdown regression with RCLError; restoring the fix passed it.
• In the prior integrated release-gate run, current/glider/USBL execution and shutdown passed 30/30 (10 each). That run included other fixes and is supporting integration evidence, not an isolated test of this PR or proof of all-platform stability.

Review / remaining scope

The separate ros_gz_bridge shutdown issue is upstream, not fixed by these guards. ARM64/AMD64 CI and maintainer review remain required before merge; an earlier ARM64 runner-space failure was infrastructure-related. Please leave this PR unmerged until reviewed.