Cleanup connect result handling

[yknoya]

Background

This PR is derived from #12729.
Since the changes in that PR mixed a bug fix and cleanup work, we decided to split them into two separate PRs.
This PR extracts and focuses on the cleanup portion.

The motivation for this cleanup is that pre-setting EIO in connect_result before a connection attempt is error-prone.
Instead, we want to set an error in connect_result only when a connection error actually occurs.
This work was triggered by the following review comment:
#12729 (comment)

Note

For the background on why EIO started being pre-set in connect_result before connection attempts, see #7580

Cleanup details

See the following comment for details:
#12729 (comment)

[bneradt]

I think this looks fine as far as I can tell. I'm a bit nervous, though, since modifications in this logic tend to result in subtle bugs. I'll build a dev rpm and try this in a production box and report back.

[bneradt]

Should this use set_fail?

[bneradt]

When I cherry-picked this locally and put this in production, it was unstable. I got the following crash:

traffic_server: received signal 11 (Segmentation fault)
traffic_server - STACK TRACE:
/lib64/libpthread.so.0(+0x12990)[0x7f575fd65990]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(HdrHeap::unmarshal(int, int, HdrHeapObjImpl**, RefCountObj*)+0x7)[0x6b3057]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(HTTPInfo::unmarshal(char*, int, RefCountObj*)+0xeb)[0x6b0d8b]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server[0x7200c9]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(CacheVC::handleReadDone(int, Event*)+0xe58)[0x7248f8]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(AIOCallback::io_complete(int, void*)+0x251)[0x745861]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(EThread::process_event(Event*, int)+0x3dc)[0x87b35c]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(EThread::process_queue(Queue<Event, Event::Link_link>*, int*, int*)+0x276)[0x87bb56]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(EThread::execute_regular()+0x147)[0x87bfd7]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server(EThread::execute()+0x1c2)[0x87cc12]
/opt/edge/2.0/trafficserver/10.0/bin/traffic_server[0x879702]
/lib64/libpthread.so.0(+0x81ca)[0x7f575fd5b1ca]
/lib64/libc.so.6(clone+0x43)[0x7f575f9b68d3]

Sadly, it's not obvious how that relates to the patch. But that's the problem with this code. We do use global origin side connection share pooling, so we make use of the ConnectingEntry logic.

[yknoya]

@bneradt
Thanks for the review and for testing this in production. I'll investigate the cause of the crash, so please give me some time.