nvDebugCheck fail with toFloat10 and toFloat11 with 0.f as input

[Zvork]

Float10 and Float11 encoding of zero (and very small float values) in toFloat10 and toFloat11 functions of CompressorRGB.cpp produce and assertion in the nvDebugCheck for the exponent field. The E variable is of type unsigned int and very small values produce a negative value of E, which when cast to unsigned gives a number greater than the 32 limit in nvDebugCheck. In this state, setting a compressor output format to R11_G11_B10F has a high chance of producing the assertion and failing.