feat: ignore trace decision messages produced by the publishers

[VinozzZ]

Which problem is this PR solving?

collect loop iteration is precious to make sure Refinery is able to process incoming data as soon as possible. Trace decision pubsub messages are broadcasted to all peers including the publisher itself. There's no need for the publisher to process those messages since it already knows about those trace decisions.

Short description of the changes

• prefix each decision messages with the peer ID. I didn't include it into the compression logic. That way we can skip the whole decompression process for a message if we know it's produced by its publisher
• ignore decision messages in the pubsub callback before we ask the collect loop to process them

[kentquirk]

It could be faster if you want.

Also, the choice of | is a good one, but it would help to include a note that it can't occur in a URL or an IP address.