Optimize RLP length calculation

[PaulusParssinen]

Changes

• Make Rlp.LengthOfLength branchless
  • This is just calculating the amount of bytes required the represent the value, we can CLZ it. This makes the method inlineable for the JIT thanks to much tighter IL (and codegen).
  • SharpLab
• Make Rlp.LengthOf almost branchless
  • We can use similar bit tricks here. Wasn't able to get rid of the branch that handles the 0 < value < 128 special case, should still be friendlier to the branch prediction and to the JIT.
  • SharpLab
• Use BinaryPrimitives in Rlp.SerializeLength
  • Little bit better codegen. We get bswap for 32-bit case and reduce remove few bounds-checks. We could also reduce bounds-checks for the 24-bit case but I thought this change more as a refactor to utilize the "new" built-in APIs.

Types of changes

What types of changes does your code introduce?

• Optimization

Testing

Requires testing

• No

Notes on testing

All tests green locally.

[benaadams]

Fails Negative_port_just_in_case_for_resilience currently; but I very much like the change

[benaadams]

This wild set of if statements has always bothered me when I saw it

[benaadams]

@PaulusParssinen thank you for your contribution