Transaction forwarding from replicas to the leader should occur through another internal RPC endpoint instead of eth_sendRawTransaction.

[arthurmm-cloudwalk]

• Inflated metrics. Today, each transaction that arrives at a replica generates two eth_sendRawTransaction metrics. 1k requests in production are merely replicas forwarding to the leader.

• Passing the client's request from the replica to the leader. Today, all these requests arrive as stratus in the client instead of having the original client, which impacts traceability.

• Error handling. Any error from the leader is interpreted as an internal error by the replica. In this scenario, the replica should simply pass on the leader's response without needing to interpret anything.

• Performance. Decoding the RLP payload of the transaction twice is slow, especially the part of calculating the transaction signer.