Switch to using ros_gz_sim to run Gazebo

[luca-della-vedova]

Bug fix

Fixed bug

Brought up in open-rmf/rmf_visualization#88 (comment).
It seems argument substitution is behaving a bit oddly with our current way to run gazebo through the executable tag.
Specifically, on Rolling the gz_headless variable seems to break the launch command, as shown in the comment (it also doesn't work for me, but running in headless mode actually works).

Fix applied

I tried to change the content of the gz_headless variable to contain the whole command, change its position in the command, various combination of content but I can't quite figure out where the issue exactly lies. It's really very odd, sometimes only headless mode works, sometimes only normal mode works, but never both at the same time.
By contrast, using ros_gz_sim seems to always work. This seems to be also the officially recommended way to launch Gazebo based on the official docs.

GenAI Use

We follow OSRA's policy on GenAI tools

I tested this on both a Jazzy and a Rolling machine.

• I used a GenAI tool in this PR.
• I did not use GenAI

[xiyuoh]

Awesome! Tested this for both Jazzy and Rolling, am able to run demos for hotel and office. Thanks for the quick fix!

[luca-della-vedova]

The CI failure is expected since the fixed rviz_common package is not part of the sync yet and not caused by this PR that doesn't touch anything related to Qt, will merge and bloom a new version