All persistent memory in the vnr module is allocated within VNR and not partly in VNR and partly provided by the user as was done earlier. This will likely change again when we think about supporting multiple models but good for now.
Changed avona::vnr::inference to INTERFACE instead of STATIC since it depends on HW target to be specified at compile time. Since lib_ic was linking against avona::vnr::inference, to avoid having to convert lib_ic into an INTERFACE lib as well, I've moved calc_vnr_pred out of IC and added it as a separate bit of shared code that is compiled as an INTERFACE library. lib_ic no longer depends on llib_vnr.
https://github.com/xmos/sw_avona/issues/311 https://github.com/xmos/sw_avona/issues/282
Changes in this PR: