Improve iridas_cube parsing speed

[remia]

While investigating long processor creation time (> 3.5sec), I noticed iridas_cube was really slow compared to spi3d for 3DLUT loading (on my system around 900ms and 250ms respectively for a 65x65x65 LUT). This brings it to similar performance.

Here is a benchmark using a 65x65x65 .cube 3DLUT, processor creation time (mean of 3 runs ocioperf --transform):

	main	pr
macOS	376 ms	216 ms
CentOS 7	927 ms	347 ms
Windows 10	1116 ms	715 ms

[remia]

Anyone knows why we use different code path for sscanf on Linux and Windows? Is there a reason (like compiler compatibility) that prevent us to use sscanf_s everywhere?

[hodoulp]

Great performance improvement.

[remia]

I'll try to include Windows numbers shortly.

Edit: that's done.

[doug-walker]

Thanks for the improvement Remi. Apologies for the delay in reviewing this.