Update readme to mention how much memory and swap is required

[ltfschoen]

In QA.md, for me to do the "Build QA engine" step. I tried just using a cheaper t3.xlarge AWS instance (i.e. 4 CPU, 16Gb RAM, 8Gb SSD, NO GPU) to see how it would perform.
First I ran out of SSD space, so I increased the SSD a bit from 8Gb to 32 Gb SSD.
Then I got an error that it didn't have a GPU, which I resolved by changing 'cuda_devices': [0, 1] in examples/pipeline/inference/config.yml to 'cuda_devices': [-1, -1] since each element in the array represents either a CPU (i.e. [-1, -1, -1, -1...]) or GPU (i.e. [0, 1, 2, 3...]), where to use CPU only you have to set each CPU to -1
Then I still ran out of RAM, so had to add 9Gb of swap space..., and it used up 50% of that swap space before I could get the "Build QA engine" step to fully work

'cuda_devices': [0, 1]