op-batcher: pending transactions is never higher than 1

[geoknee]

We are seeing this in production networks. With the current way the batcher works, it should send transactions asynchronously (but with nonces determined synchronously).

We would therefore expect (particularly in times of high thoughput) that there would be multiple pending transactions (up to the configured limit, typically 5 or so).

There could be a bug in the metrics, or the transactions are not actually being handled in the expected way (which would indicate that we are hamstringing the batcher's throughput unintentionally).