Better recovery from `preconfBlock` timeout

Right now, when we get Failed to call driver RPC for API 'preconfBlocks' within the duration (1000ms), we remove the block that failed to preconfBlock like this:

        match self
            .taiko
            .advance_head_to_new_l2_block(
                l2_block,
                anchor_block_id,
                l2_slot_info,
                end_of_sequencing,
            )
            .await
        {
            Ok(preconfed_block) => Ok(preconfed_block),
            Err(err) => {
                error!("Failed to advance head to new L2 block: {}", err);
                self.remove_last_l2_block().await?;
                Ok(None)
            }
        }

However, this sometimes causes the node to be stuck as the preconfBlock may actually be reflected in taiko geth, but removed from side car. This causes the side car to be stuck with highestUnsafeL2PayloadBlockID: 407, different from Taiko Geth Height: 408.

We should handle recovery from such time outs in a better way.

Start with the shown preconfBlock timeout handling and the call to remove_last_l2_block(). Trace how the sidecar state is reconciled with Taiko Geth after the timeout, especially the highestUnsafeL2PayloadBlockID mismatch. Done means a timeout recovery path no longer leaves the sidecar behind when the preconfBlock was accepted by Taiko Geth.