Change the memory allocator of the standalone kernel

[tomaka]

At the moment we use the linked-list-allocator crate.
While it works fine, it is a very naive allocator and clearly isn't meant to be used when the load of allocations/deallocations is heavy.

Additionally, the allocator is single-threaded and is wrapped around a Spinlock. Multiple CPUs cannot at the moment allocate memory in parallel.

It would be rather hard to measure how much time the kernel spends inside of the memory allocator, but I can totally imagine a world where it is the allocator that is currently the biggest choke point.

[tomaka]

It would be rather hard to measure how much time the kernel spends inside of the memory allocator,

It's actually not that hard. This patch shows that the median time for a memory allocation or deallocation seems to be around 256µs on QEmu.

[tomaka]

Note that 256µs remains a lot even from within an emulator. The time malloc takes on a desktop computer is more in the range of the hundreds of nanoseconds.