Save RAM

[fulmicoton]

Right now the independant VecDeque are taking too much memory.

Our logic to avoid resizing/reallocating VecDeque continuously is working well, which means that all queues are taking roughly the needed memory at full capacity.
That's twice the capacity in use.

We went for VecDeque for its capacity to offer large slab of data.
We could save this half by using a different datastructure.

For instance a simple unrolled linked list would do the job.