by default the topic names are flattened so that sensor data doesn't move with the tf transforms (instead it is fixed at the time of capture); this could be changed by explicitlity mapping the sensor topics to a child of the tf entity path
the tftree has to be specified in advance (otherwise it would also have to be flattened, which wouldn't go well together with the URDF loader)
the incoming raw tf messages are immediately logged to Rerun; in addition interpolated transforms from tf are logged with a delay (the assumption is that after this delay the whole tree has been updated); this way we get the latest poses without any delays, but it might not look smooth in real-time; in the playback or when looking at the data with at least the same delay, we get the smooth interpolated transform
Requires the prerelease. Maybe best not to merge before 0.15 is released.
Adds support for:
Adds new Spot example that illustrates these new features
https://github.com/rerun-io/cpp-example-ros-bridge/assets/9785832/5972d7c9-a5a5-4cb8-8c3e-f76e29e7d047
Some decisions I made that might be up for debate