If the main request was finalized while a background request performed an
asynchronous operation, the main request ended up in ngx_http_writer() and was
not finalized until a network event or a timeout.  For example, cache
background update with aio enabled made nginx unable to process further client
requests or close the connection, keeping it open until client closes it.

Now regular finalization of the main request is not suspended because of an
asynchronous operation in another request.

If a background request was terminated while an asynchronous operation was in
progress, background request's write event handler was changed to
ngx_http_request_finalizer() and never called again.

Now, whenever a request is terminated while an asynchronous operation is in
progress, connection error flag is set to make further finalizations of any
request with this connection lead to termination.

These issues appeared in 1aeaae6e9446 (not yet released).

With OpenSSL 1.1.0+, the workaround for handshake buffer size as introduced
in a720f0b0e083 (ticket #413) no longer works, as OpenSSL no longer exposes
handshake buffers, see https://github.com/openssl/openssl/commit/2e7dc7cd688.
Moreover, it is no longer possible to adjust handshake buffers at all now.

To avoid additional RTT if handshake uses more than 4k we now set TCP_NODELAY
on SSL connections before handshake.  While this still results in sub-optimal
network utilization due to incomplete packets being sent, it seems to be
better than nothing.

Previously, cache background update might not work as expected, making client
wait for it to complete before receiving the final part of a stale response.
This could happen if the response could not be sent to the client socket in one
filter chain call.

Now background cache update is done in a background subrequest.  This type of
subrequest does not block any other subrequests or the main request.

If allocation of cleanup handler in the HTTP/2 header filter failed, then
a stream might be freed with a HEADERS frame left in the output queue.

Now the HEADERS frame is accounted in the queue before trying to allocate
the cleanup handler.

Abnormally exited workers may leave locked cache entries, this can
result in the cache size on disk exceeding max_size and shared memory
exhaustion.

This change mitigates the issue by ignoring locked entries during forced
expire.  It also increases the visibility of the problem by logging such
entries.

Previously, an allocation error resulted in uninitialized memory access
when evaluating $upstream_http_ variables.

On a related note, see r->headers_out.headers cleanup work in 0cdee26605f3.

Found by gcc7 (-Wimplicit-fallthrough).