Investigate and address certain performance problems for high-volume workloads.

[alexbarcelo]

We detected certain misbehaviour when executing numerical applications without active features --i.e. doing attribute getters and attribute setters with big numerical data structures.

At first we assumed that it was some performance penalty at the gRPC or socket/serialization layer, and certain preliminary results seemed to confirm that, but after more in-depth tests the results came up unconclusive.

[alexbarcelo]

Some experiments backing up the "unconclusiveness":

(size is in kiB. Goes from 1kiB to 1GiB)

The normalized time seams to plateau, which is an expected well-behaviour. Small rpc have a big overhead, but that can be assumed. For these experiments the base rpc overhead is around $10^{-3}s$.

[Note: The difference between ndarray and numerical_array is implemented in bsc-dom/pyclay/pull/59]