Clean RLP APIs

[tkstanczak]

We would like to have RLP as a relatively independent library (either as a single project or coupled with Nethermind.Core only).

At the moment there are numerous APIs in RLP to allow for greater code optimization in Patricia Tree and other areas. As a result, the APIs can be confusing and not easily reusable in other projects.

Any changes here would need to assure that the existing optimizations are still supported.

In general, we have three approaches now:

1. original that does quite a lot of allocations
2. new one with creating buffers and block copy
3. newer one with spans

They also take different arguments depending on the level of optimizations needed (with Patricia Tree going most hardcore).

It would be great to review, improve and systematize the RLP APIs.