Fix material changes within an object being ignored

[Skynoodle]

Previously, a single object would always create a single model with the
last specified material. A usemtl statement in the middle of an object
will now emit a model at that point (with the previous material).

[Twinklebear]

It looks like the cornell box tests are failing, my guess is this is due to the stray usemtl white which is repeated in the obj file https://github.com/Twinklebear/tobj/blob/master/cornell_box.obj#L122 . With this change we'd now output a separate model at that point when we probably shouldn't (since it's the same material), so the test gets 9 models instead of the expected 8 https://github.com/Twinklebear/tobj/blob/master/tests/lib.rs#L250.

I think in the case of this OBJ file (which is sort of not well formed) if we get a repeated tag of the currently active material we should just ignore it.

[Skynoodle]

Whoops! Yeah, that sounds likely - it just naively emits a new model even if the usemtl is a repeat. Happy to fix that up, if the basic approach is acceptable (I noticed there was a discussion about this vs emitting materials per-primitive in the related issue).

[Twinklebear]

I think for the OBJ loader it's probably best to emit a model per-material, instead of per-primitive material IDs that a renderer would use, which is probably best left to the application.

Thanks @Skynoodle , this looks good!