[HTTP/2] BlocksQueue can probably be optimized

[c410-f3r]

BlocksQueue is a fundamental CSR-inspired data structure for HpackHeaders and any optimization will likely benefit HTTP/2.

It is a circular buffer like VecDeque but instead of storing individual elements, it stores chunks of elements that are logically demarcated. Another important feature of BlocksQueue is the fact that chunks never intersect boundaries.

Three "push_front" operations. `H` means head.

// One block. First has 1 element
Buffer: [. . . . . H . .]

// Two blocks. First has 1 element and second has 3 elements
Buffer: [. . H * * * . .]

// Three blocks. First has 1 element, second has 3 elements and third has 3 elements
// To avoid an intersection, the buffer had to be increased with shifted elements.
Buffer: [. . H * * * * * *]

I am not totally sure but the implementation can probably be optimized to generate more efficient assembly, specially in the "push_front" method.