This PR is essentially the same as #1632, which has been invalidated by some black magic.
This PR heavily improves the performance on FlatBuffer by introducing cache mechanisms on indexed access.
This is possible through removal of byte-level manipulation on Buffers and re-introduction of a constant-time length attribute on Buffer.
nitmd is now practicable again and should no longer take too much time when running benches.
This PR still depends on #1628, and as such will not pass no tests as long as it is not integrated
Pull-Request: #1644