Maybe a better approach altogether is to drop the this.headers on the client altogether, and simply have each
endpoint add its own headers. The endpoint is the thing that knows how the body should be encoded, so it probably makes more sense to do it there.

I think we'd still need this.headers in order to manage things like user agent.

The tricky thing with managing this on a per endpoint basis is that application/json is a sensible default, and the desired behaviour is that the Content-Type header is unset, rather than overridden, as setting the boundary happens at the fetch level, so even if setting the headers was moved to the endpoint, there would still be an issue.

Keen to hear your thoughts though, I may be missing the mark!