Fix NaN in friction calculations for collision pairs

[Tanisha-Sharma005]

Problem

Friction calculations in collision_pair.cpp produced NaN values due to improper handling of collision pair types and geom data arrays.

Solution

• Added proper validation for wp_array_dyn_t friction pair types
• Fixed geom data array processing to prevent uninitialized reads
• Ensured friction coefficients are properly initialized before use

Testing

Verified friction calculations now produce valid float values in all collision scenarios.

Closes #942

[thowell]

@Tanisha-Sharma005 thank you for your contribution to mujoco warp! the nans reported in #942 are likely the result of floating point precision and not a bug in the implementation. if you are interested in further contributing to mujoco warp, please comment on an issue marked with 'good first issue'. thanks!