Add GPU usage

[dhiltgen]

Help users understand how much of the model fit into their GPU without having to resort to inspecting the server log

A few examples from different systems and models

eval rate:            4.40 tokens/s
gpu usage:            1 GPU (14/27 layers) 3.2 GB (2.0 GB GPU)

eval rate:            6.64 tokens/s
gpu usage:            1 GPU (27/27 layers) 3.2 GB

eval rate:            18.44 tokens/s
gpu usage:            2 GPUs (27/33 layers) 27 GB (24 GB GPU)

eval rate:            19.58 tokens/s
gpu usage:            CPU (0/27 layers) 3.2 GB

[mxyng]

#4190 broke lint on windows. gofmt is still a problem

[dhiltgen]

Still chewing on the optimal UX, so I've removed the UX from this PR to lay the groundwork for a follow up PR to expose it in the UX.