The human hand is not sensitive enough to detect which side of a particular die is heaviest. Otherwise we wouldn't need other ways of testing dice.
It's not a matter of human hand bias (even though you do pick non-randomly from the bag). It's a matter of dice bias. As I said, if there's 50% chance to give 6, then it'll mostly give sixes, whether you use a bag in which you scramble them a hundred times, or not.
cakewallet had a csprng in the code the problem was it also had a fallback which kicked in if the csprng failed to return a seed.
I don't know what's cakewallet, if it's open-source, if it's peer reviewed, if it's a Bitcoin wallet etc. Would you mind sharing a link that describes the CSPRNG failure in that software? As far as open-source, reputable Bitcoin wallet software are concerned, such as Electrum, there has never been such case.