I think the !body_format && body is pretty clear. What should probably be clarified is:

• body_format && body, and in this case I assume body_format will take precedence (please also update the comment of body - see example). (after briefly looking at the code changes, it is not clear to me if this requires the body to be present, and works on top of that or not - this should be clarified).
• !body_format && !body, no body will be in the generated response.

I copied the definition from the custom_response_filter - but I can definitely make this clearer. This should be updated within that as well, but that's out of scope for this PR.

OOC what's the benefit of having a function call here?
If the code is not reused, IMHO it should just be here.
(same applies to the test below)

I thought that too, but this is how it's already structured elsewhere in this file, and I try not to ask for cleaning up other people's mess, or being inconsistent with it.
(I was thinking I might clean it up afterwards.)

I thought that the testDirectResponseBodySize() method is being reused in some tests, and that's why it was made a function. Granted, there may be a better approach here.
Specifically the new functions don't seem to be reused, so IMHO the additional dereference just makes it a bit more complex to understand.

The whole subfunction thing is part of my pet testing peeve, that if you do a subfunction like that and the test fails, the output you get is frequently much less helpful than it would be if you did the helper functions differently.
e.g. instead of

testDirectResponseBodySize(1);

you could structure it

auto body = getDirectResponseBody();
EXPECT_THAT(body, HasLength(1));

which is very slightly longer but significantly clearer and makes any test failure occur in the test function, and also can log more useful information on failure (e.g. that the body is "banana" rather than just that the body's length was not 1.)

Anyway, for the purposes of this PR my point is just that I'm recommending not fixing this in this PR, because there were bad precedents, and I'll clean it up better afterwards. :)

testDirectResponseFile() is also not being reused anywhere else, but was also made into a function.