op-stack EL: publish p2p port + keep discovery on for node_sync nodes (49 composes)

Operator decision 2026-10-04 (sebastian): consensus-layer-synced op-stack nodes used to run
with --nodiscover / --disable-discovery and no published p2p port. A 0-peer EL can never
backfill a dropped-gossip gap, so op-node wedges in FCU-SYNCING until L1 derivation
catches up (katana 09-19..26, zircuit, mode). Blocks still arrive only via op-node; EL
peers are for gap-fill. Our own copies now discover each other implicitly.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RzxemENdcZBCxdGrzXiFvS

op/reth/katana-mainnet-op-reth-archive-trace.yml

@@ -46,6 +46,9 @@ services:
       nofile: 1048576    # Max open files (for RPC/WS connections)
       memlock: -1        # Disable memory locking limits (for in-memory DBs like MDBX)
     user: root
+    ports:
+      - 14265:14265
+      - 14265:14265/udp
     expose:
       - 8545
       - 9001
@@ -58,11 +61,16 @@ services:
       - --chain=/config/genesis.json
       - --config=/config/reth/reth.toml
       - --datadir=/root/.local/share/reth
-      - --disable-discovery
+      - --discovery.addr=0.0.0.0
+      - --discovery.port=14265
+      - --discovery.v5.addr=0.0.0.0
+      - --discovery.v5.port=14265
       - --engine.cross-block-cache-size=${KATANA_MAINNET_RETH_STATE_CACHE:-4096}
       - --max-inbound-peers=50
       - --max-outbound-peers=50
       - --metrics=0.0.0.0:9001
+      - --nat=extip:${IP}
+      - --port=14265
       - --rollup.sequencer-http=https://rpc.katana.network
       - --rpc-cache.max-blocks=10000
       - --rpc-cache.max-concurrent-db-requests=2048