Submit FS#11365. Speed up mp3 decoding on ARM processors. It is faster to use the...

Submit FS#11365. Speed up mp3 decoding on ARM processors. It is faster to use the C-implementation of dct32 compiled with -O1 as the asm implementation of dct32 compiled with -O2. Configuration for Coldfire processors is untouched. In the new configuration the stack of the COP decoding thread needs to be increased on dualcore targets. Speed is increasing by up to 0.9 MHz (-O2 against -O1 using eabi).

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@26746 a1c6a512-1295-4272-9138-f99709370657