Possible nan in EVS IGF decoder

In the function IGF_replaceTCXNoise_2 (only used in EVS) a possible nan value can occur. This function is called for every IGF tile with no or mid-whitening. If there is only one bin in the tile which is classified as noise and own_rand function happens to give back 0 for this one bin, then a division by FLT_MIN occurs which can result in nan.

This scenario should be extremely rare (it's even less likely than the corresponding problem in IVAS due to larger tile sizes in the EVS IGF configs) but can destroy the whole frame if it does happen and should therefore be fixed.