[Lut] correct 3D Lut trilinear algorithm

This fixes
- The indices to get 8 corner points are not correct.
- The interpolation calculation is also wrong

The order to interpolate also matters. We flatten original[x][y][z] to
flat[x * size * size + y * size + z], therefore, the order to
interpolate is z, y and x

Bug: 404391247
Bug: 333088569
Test: test on silkfx
Flag: EXEMPT bugfix

Change-Id: Ifbeeb59b656fdb33c1cb4ff65cc542882804636e