op-node: instability triggered by L1 RPC delays

Initially reported by Alchemy.

* Cloudflare has had a bunch of issues -> unreliable L1 connection on both beaconchain and L1 RPCs
* OP Stack has trouble producing unsafe blocks when the L1 RPC is slow
* Any time you exceed .5 second delay, some processes (derivation) gets stuck. This can cause chain halts, pretty reliably
* Alchemy has been handling this on their end via proxyd, but we may be able to fix this in-protocol
* Can find this by introducing an artificial .5 lag in our testing

May fit into the op-node v2 refactors.

Start in the op-node area by reproducing the issue with an artificial 0.5-second L1 RPC delay, then trace how derivation and unsafe block production respond to the delayed connection. Done means the op-node remains able to produce unsafe blocks and avoids chain halts under the reported L1 RPC and beaconchain delays.

Written by the indexing model from the issue text.