Malformed chunked response detector creates false-positive ‘premature close’ errors when HTTP Agent involved about node-fetch

[Yogu]

I needed to debug the issue in order to understand this issue, but I think I do now. Just posting this comment to re-phrase the issue in my words, maybe it helps someone else. You mentioned it in #1576 with this comment

If you folks are using an http.Agent with your connections in keepAlive mode with a fixed number of maxSockets, #1767 is the problem.

and this actually means "if these conditions are met, you will always encounter the issue" (so this issue here) regardless of chunk/content lengths (what was discussed in #1576)

The reasons seems to be: node-fetch assumes that the body 'data' event fires before response 'close' event. This is not true when an agent wants to reuse a connection where the request already completed, but the consumer has not read the body yet. Seems like node does not wait for the consumer to actually consume the response body and instead reads and buffers it.