Update to latest Magnum with full skinning support and much more

[mosra]

Motivation and Context

Upgrades Magnum to latest, containing the following:

• builtin joint ID and weight mesh attributes, including support in MeshTools::compile()
• skinning support in the PhongGL, FlatGL and MeshVisualizerGL shaders, including instancing and multi-draw variants
• import of builtin skinning attributes in AssimpImporter and GltfImporter
• new MaterialTools library which will gradually receive features for material simplification, deduplication and conversion between Phong and PBR representation; and a new DebugTools::CompareMaterial utility for convenient testing of material conversion and processing code
• minor user convenience improvements like a new sliceSize() API in all array, string and view classes accepting a pair of offset+size instead of begin+end
• more forgiving GltfImporter behavior with unsupported vertex formats, now it imports the attributes as custom instead of failing the whole mesh import
• PngImporter and PngImageConverter now handle libpng upgrades more robustly, failing only in case of an ABI break
• ability to disable upsampling in StbResizeImageConverter, useful with batch texture resizing conversions
• support for string fields in SceneData together with importing string "extras" from glTF nodes (such as various Floorplanner tags, shown bellow), essential also for efficient, SceneData-backed binary serialization format for gfx-replay (what's still missing is efficient bitflag representation, that's coming next)

Builtin shader constructors got changed to accept a Configuration class because with skinning the argument count was getting out of hand. While the previous constructor signatures are still available, they're marked as deprecated so I updated the code to use the new APIs.

Skinning support and skinned animation playback in the magnum-player utility and its prebuilt binaries, as I advertised on the call today, will arrive in the next days.

How Has This Been Tested

My CIs pass.

[mosra]

Anybody knows what's up with the CI? Both of the failed jobs end with a cascade of errors that have

AttributeError: module 'torch._C' has no attribute '_cuda_setDevice'

at the top. I was getting this with #1947 and thought the reason was that it was made from a fork, because if I pushed the same changes to a branch on the main repo, the CI passed. But this PR somehow doesn't.

[Skylion007]

@mosra Pull from master that should have been fixed. It's pulling the CPU only version of Torch I think.

[mosra]

But this PR is based off current main, commit fffa543 from a week ago...

[Skylion007]

Oh boy, did it break again? Ping @aclegg3

[0mdc]

LGTM!