Experimental EVM mempool not gossiping nonce + n transactions

[jhelison]

Issue description

Recently, we have found an issue where the transactions with nonce + n were not being gossiped.
This is the behaviour that we have faced:

If sent to a Validator node with enough voting power

Within a validator, the message is kept on the local mempool and all TXs are added to a block:

1. TX A with nonce + 1 is created, lands in queue
2. TX B with nonce + 0 is created, lands in runnable
3. TX B is processed and lands in a block
4. TX A is processed and lands in a block

Non validator node

On non-validator nodes, only the runnable TX is gossiped to the validator and added to a block:

1. TX A with nonce + 1 is created, lands in queue
2. TX B with nonce + 0 is created, lands in runnable
3. TX B is processed and lands in a block
4. TX A is now stuck and can't be re-sent unless overpriced under the same nonce

How to replicate

This issue can be replicated by:

• Have a node with the Experimental EVM Mempool enabled
• A node available with P2P connections to a validator, transactions must be sent to the public node

The following must be applied:

1. Send an EVM TX with nonce + 1
2. Send an EVM TX with nonce + 0 (expected nonce)

The following will happen:

• Only the TX with nonce + 0 will be executed
• TX with nonce + 1 will be stuck on the local mempool and will not be gossiped
• Wallet will be stuck unless an overpriced TX is sent to recover the Nonce + 1 TX

The following would be the expected behaviour:

• Nonce + 0 should be executed
• Nonce + 1 should be executed (after internal mempool reordering)

[linear]

EVM-403

[jhelison]

This is a proposal to fix the issue #890