reverseproxy: Don't buffer chunked requests (fix #5366)

[mholt]

Mostly reverts 845bc4d (#5289) -- that was my bad, totally an oversight on my part of the code review.

Adds warning for unsafe config. The buffer_requests and buffer_responses config params, together with a separate max_buffer_size parameter, were not safely designed: very prone to configuring unbounded buffers. I'm not sure why I ever thought that was fine.

Deprecates unsafe properties in favor of simpler, safer designed ones.

We'll need a better fix for #5236 -- even if it's the same kind of fix (buffering, and resetting the Content-Length) as long as it's opt-in this time.

/cc @u5surf What do you think?

I think this patch merits a quick release of v2.6.4.

[mholt]

I could remove isChunkedRequest() to satisfy the linter, but I left it in there because I want to use it to apply @u5surf's earlier patch, just in a more careful/opt-in way. I am just not sure how to go about that. Would love any ideas/insights!

[francislavoie]

Seems good to me.

I think we can just deleted the chunked encoding code. If users who that fix helped find that they need buffering, they can enable it for themselves (and now, set limits).

[u5surf]

@mholt
Sorry for late.
I also agree with not to buffering such a large chunked contents. I lost my sight at this point because I misunderstood that it has already taken this problem in h.bufferedBody or somewhere.
Your patch looks good to me. Thank
you.

[mholt]

@u5surf No worries, thank you for checking it out! And thank you for the original patch. The mistake was my oversight. Hope you'll contribute again.

[u5surf]

@mholt
I have left a few comment, I still might have confused for me the new configurations.

[mholt]

@u5surf Ah wow, thank you for your review -- I had a lot going on yesterday and looks like I got distracted during a search-replace and mixed them up 🙈

Thank you for catching those errors! I've pushed what I hope is the final change for this patch.