Treating initial mesh calculation as fallible

[bellwether-softworks]

I'm attempting to naïvely perform convex hull transformations to convert point clouds to meshes, but frequently encounter a shape that results in a panic. It would be helpful to know in advance that a shape isn't regarded as valid from Parry's perspective, so to support this I'm handling potential points of failure with an error struct. The original get_initial_mesh function now calls this modified method with an .unwrap() call, thus preserving the original function contract.

:bulb: Ideally, the convex hull transformation would similarly allow error handling, but this serves as an interim solution.

I've also added a test for a point cloud known to result in failure in an f32 environment.

Edit: I've also exposed some internals to allow for external consumers to take advantage of this behavior (in particular, as can be seen in the test, a consumer needs access to normalize as well as try_get_initial_mesh). My approach is admittedly heavy-handed, and I can revisit this as necessary.

[sebcrozet]

Hey thank you for this PR! It took me forever to get to it, but it can be merged now. I changed it to make the 3D convex-hull calculation itself faillible (with try_convex_hull) instead of exposing internals to the user.