fix: preserve subsecond timestamps in derived depth messages

[yeseorizi]

Stacked on #15. The base is fix/ros-plugin-shutdown-20261008 so the review diff contains only the depth timestamp fix and its test. Retarget to main after #15 is merged, preserving this small diff.

Problem

The pressure message includes seconds and nanoseconds, but its derived depth message copied only whole seconds from the measurement time. For example, measurements at 0.125 s and 0.325 s both produced a depth timestamp of zero.

Fix

Copy the pressure message's full header to the depth message generated from that same measurement. Pressure, variance, depth values and topic names are unchanged.

Validation

• New ROS/Gazebo plugin regression receives both messages at 125 ms and 325 ms, checks identical timestamps, and checks the existing pressure, variance and depth values.
• The isolated timestamp change builds and passes the timestamp regression plus five camera tests on Linux ARM64 / ROS 2 Lyrical / Gazebo Jetty.
• The final stack with fix: guard ROS plugin updates during context shutdown #15 passed all seven cases (five camera, one pressure shutdown, one timestamp). Replacing the pressure implementation with unchanged main failed both new pressure regressions; restoring the fix passed all seven again.

This is message-timestamp consistency, not a claim of physical pressure-sensor accuracy or all-platform validation. No merge is requested before CI and review.