EVS stereo panning in JBM is suboptimal

This is a follow up to the discussion at !2346 (comment 96579).

The EVS stereo panning feature in the current framework works as follows:

• mono audio from decoder -> panning into stereo audio -> JBM is done on the two stereo channels

This is obviously suboptimal and the following changes could be done:

• do the panning only after the JBM, i.e. mono audio from decoder -> JBM -> panning into stereo audio
• JBM would be run on only one channel which would save complexity and memory of one APA channel buffer
• in BASOP, the tmp_apaExecBuffer[] buffer in apa_exec_evs_wrapper() could be shortened to one channel only which would again save stack memory

Note that it is currently the APA scaling routine that is on the worst stack case in the EVS decoder.