Fix locale-dependent string-to-float conversion when loading Collada meshes

[peci1]

Description

Fixes #1665.

<unit meter=""> attribute of of Collada meshes was read in a locale-dependent manner. This PR fixes it.

Checklist

• If you are addressing rendering issues, please provide:
  • Images of both, broken and fixed renderings. Marker load mesh resource success, but not show in window? #1665 (comment)
  • Source code to reproduce the issue, e.g. a YAML or rosbag file with a MarkerArray msg. Marker load mesh resource success, but not show in window? #1665 (comment)
• If you are changing GUI, please include screenshots showing how things looked before and after.
• Choose the proper target branch: latest release branch, for non-ABI-breaking changes, future release branch otherwise.
  Due to the lack of active maintainers, we cannot provide support for older release branches anymore.
• Did you change how RViz works? Added new functionality? Do not forget to update the tutorials and/or documentation on the ROS wiki
• While waiting for someone to review your request, please consider reviewing another open pull request to support the maintainers of RViz. Refer to the RViz Wiki for reviewing guidelines.

[simonschmeisser]

I was also wondering if there is a function in assimp to get the scale/unit which might possible be not-broken? Then we could kill that Collada special casing and have scale support for any other file type that supports it as well (I'm sure there are some ...)

[rhaschke]

The unit scale is actually read by assimp and applied. Not sure though, why the scaling isn't actually considered. I found the post-processing option aiProcess_GlobalScale which wasn't yet used by rviz. However, it doesn't change the result 😞

[rhaschke]

Interestingly, I get these errors on your meshes, which is confusing as the XML is perfectly valid:
Collada: file:///tmp/meshes/cube.dae - Expected end of <author> element.

I'm using assimp 5.0.1~ds0-1build1 on Ubuntu 20.04. Which version do you use?

[peci1]

I'm on 4.1.0~dfsg-3 (Melodic / 18.04)

[rhaschke]

Closing in favor of #1685.