Support 32-bit float for decompressed KTX output

[solidpixel]

User would like 32-bit float data in decompressed KTX files to make the data easier to consumer in other diagnostic software.

Today we always use fp16, which is the notional decompression precision defined by the spec. To support alternative options we'd need to add an additional command line options to specify the output format.

Split off from #352