Keep large HTTP transfer payloads out of Duroxide history

Currently, large HTTP payload bytes end up in the Duroxide history. In theory, (I think) we only need to persist the things necessary to reproduce whatever orchestration decisions are necessary, and resume work. That means we shouldn't need to persist HTTP payloads, and doing so causes actual problems (performance, security, data retention, data control, etc).

It's possible that some situations will require keeping the body in the history (particularly where the bodies are small, replay/orchestration needs the body's exact value, it contains no credentials/secrets/potential PII/otherwise sensitive info, ...). That's probably fine. The precise set of conditions when these need to be kept needs to be hammered out as part of this work.

This needs some further design too.

No files, tests, or entry points are named. Start by mapping where HTTP payloads enter and are written to Duroxide history, then define which payload conditions are needed for replay and orchestration. Done means large or sensitive payloads are kept out of history while any explicitly justified small payloads remain available when required.