Wrong parameter shift in DFT Stereo adaptive stereo filling if input is 48 kHz

If hStereoDft->reverb_flag is triggered via adaptive stereo filling in the DFT Stereo encoder 3 of the residual prediction gains are not transmitted, but predicted at the decoder. For low band resolution at 13.2 and 16.4 kbps these are always the 3 highest transmitted bands. However, for high band resolution at 24.4 and 32 kbps it's the 3 bands below the 8 kHz WB border that are predicted. For the highband 2 more values are regularly transmitted in case bwidth is more than WB. These 2 parameters are then shifted down by 3 before entering the stereo parameter entropy coder.

This doesn't seem to always work correctly, though. If the input sampling rate is 48 kHz (and no lower bwidth is specified via commandline) the shift will be wrong, because instead of shifting bands 11 and 12 to 8 and 9, it will shift bands 12 and 13 (which only exists in FB and isn't even used for stereo filling) to 9 and 10. This causes the decoder to go out-of-sync and leads to using the wrong parameters for some bands.