encodeBody: "manual" is ignored on gzip

Starting very recently, Cloudflare cache appears to ignore the 'encodeBody': "manual" instruction and instead serves re-encoded files when the requester accepts only gzip encoded content (more accurately, when the requester doesn’t accept brotli).

This problem doesn’t happen in every POP. I’m able to observe this SJC, but doesn’t happen for instance in IST.

How to reproduce

curl https://kimlikdao.org -H "accept-encoding: gzip" --output index.html.gz
wc index.html.gz

The corresponding worker always caches requests with 'encodeBody': manual so one should expect files of a fixed size. However when hit SJC, different file sized can be observed (probably on-the-fly compression)

Actually, we’re able to observe this on IST as well.

Adding ‘no-transform’ to cached Response ‘cache-control’ header seems to fix this problem for now. Why is this needed though?