Fix normals Float32 precision loss in JS

[robert-salata-hl]

For some specific geometries, o3dgc.js parser fails due to inaccurate normals computations. The m_normals array was defined as Float32Array, which does not provide precision necessary for computations performed inside ComputeNormals function, later resulting in error CubeToSphere[nin.m_z] is not a function. The issue can be observed on attached geometry mesh.o3dgc. The same geometry is parsed correctly by c++ parser.
mesh.zip

Changing Float32Array to Int32Array resolves the issue, and is aligned with c++ implementation - see Open3DGC/src/o3dgc_decode_lib/inc/o3dgcSC3DMCDecoder.inl file, ProcessNormals function. The n0 variable stores the same normal values as m_normals array in js, and is declared as Vec3<long>, not Vec3<float>.