Improve PubSub resilience by switching to SubscriberClient

[danielwinkler]

This is a follow up for the ticket #1648 and #1700

• The tests are passing, let me know if you like the general approach;
• The major difference is that the ACKing is not done explicitly any more, but through the SubscriberClient API.
• The benefits of the SubscriberClient are discussed in the other tickets already.

[danielwinkler]

I just noticed that you started working on this topic in parallel, so i guess this is somewhat superfluous...

[jeremydmiller]

Ugh, sorry, bad timing. I basically started a near rewrite for V5 -- which won't quite make it in time. Let me think about it. I haven't pushed my changes yet.

[danielwinkler]

No worries at all,

it might be a better idea to go for the fundamental rewrite rather than my making it work approach.

[jeremydmiller]

@danielwinkler I'm going to take this in, but do another round of changes to it to make it a bit more idiomatic. I think "actually working" trumps "somewhat improved"

I might make a benevolent dictator decision to simplify this a bit and eliminate the "Buffered" support so we can get away w/ only having one single listener

And thank you for taking this on!

[danielwinkler]

Thanks for the feedback @jeremydmiller, glad you can make use of it.

I haven't yet looked into, but I wonder if PublisherServiceApiClient is having the same issues as SubscriberServiceApiClient, thus moving to the higher level PublisherClient on the senders side.