Remove contention on the Gic

[dgageot]

Instead of locking a single Gic, we move the contention down to the vCPU level. This gives us better performance for CPU bound tasks.

On a 8 cores M3:

• Building the Linux Kernel: 7m21s -> 6:58s
• Building Wasmer: 3m13s -> 2m30s

[dgageot]

Hi! I've pushed a fix for the linter. Can someone re-trigger the Code Quality jobs, so that I check if something else is missing? Thanks!