tr_model_iqm: rewrite NaN in model IQM input

[illwieckz]

Rewrite NaN in model IQM input.

The prifle_view.iqm model has some NaN in the input data:

• The prifle_view.iqm model has some NaN in the input data Unvanquished/Unvanquished#3252

When such model file is used, it is impossible to get anything useful when trapping for invalid operations because the first call of R_TBNtoQtangents() when loading the model will stop the execution.

There are other floats in the IQM file, we may want to not hack BuildTangents() and sanitize all the floats in the future with a generic function, but this works for now.

Here is the IQE text equivalent of the IQM format, there are many float fields like that:

pq 0 0 0 0.707107008 0.000172803004 0.707107008 -0

vp -7.02030039 -5.93917179 -10.5407791
vn -0.373933047 -0.803014815 -0.464048833
vx 0.618980289 0.156532541 -0.769649923 -1
vt 0.857393026 0.983985007
vb 9 1

Here is an example of broken input, the current code currently only cares about vx.

vp -4.99237871 -17.1318531 -6.18218517
vn -0.0307731722 8.04468016e-07 -0.999526381
vx nan nan nan 1
vt 0.548720002 0.101943001
vb 1 1

[VReaperV]

I assume this doesn't have much of an effect on loading times?

There are other floats in the IQM file, we may want to not hack BuildTangents() and sanitize all the floats in the future ith a generic function, but this work for now.

This should be added as a comment.