GPU recon doesn't check for numeric overflow

[Jeremy Gill]

While converting internal floating-point data to 8-bit or 16-bit integer, one must check for numerical overflow and clamp appropriately - it's better to clamp and saturate, rather than have underflow and overflow.

[Jeremy Gill]

Addressed by commit :

68a6ede229303b7c2369e512acaab70436f049ef