Memory optimization for RPC Jumpdests fetching

[einar-polygon]

This continues the work in #427 and addresses one problem: CREATE and CREATE2 required memory dumps to obtain the relevant contract code and code_hash for call stack simulation, and we opted instead to fail over to simulation, since memory dumps were slow.

This branch introduces the following changes:

• Reinstantiate the match case for CREATE and CREATE2, but ignore code and code_hash.
• Prove only the current context. The old version generates all Jumpdest proofs in the very first context. This meant that the witness had to provide the relevant mapping between code_hash and context. This change means two things: current_code and current_code_hash is always available, so we do away with the mapping in the witness, essentially simplifying it to a mapping ctx->[offset].
• Simplify the witness including removal of the tracking of code and code_hash in the Jumpdest generation.

Work has started in PR: Prefetch JUMPDESTs through RPC with progressive proving #765

Status: The first 100 blocks on testchain2 succeed, but there are still many errors later.