Fix unobserved SocketException (rebase of PR 771, which addresses Issue 770)

[gbirchmeier]

resolves #770 (original issue)
resolves #771 (original PR, this is a rebase)

Perform socket read operations according to Task-based asynchronous pattern (TAP) instead of Asynchronous Programming Model (APM)

[gbirchmeier]

@nmandzyk and @Rob-Hague -

I rebased the original PR #771 on latest. SocketReader now has null-analysis turned on, and the compiler is flagging that _readCancellationTokenSource can be null at the start of ReadSome().

Looking into that has led me to DisconnectClient(), which... well, is there a reason to assign _readCancelationTokenSource and _currentReadTask to null here? If we left _readCancellationTokenSource as a disposed non-null object, the logic wouldn't change, and then we could take the ? off the var declaration so it would never be null and the warning in ReadSome() would go away.

(If I do this, and by some weird chance the var gets referenced after DisconnectClient(), then we'd now get a ObjectDisposedException instead of a NullRefException. Barely a difference functionally, and probably a little easier to diagnose if we need to follow up on it).

(I suspect SocketInitiatorThread might have a similar issue, but I'll enable nullable on this file and clean it up in another commit.)

[gbirchmeier]

I've made the nullable fix in the second commit. Please tell me what you think.

[Rob-Hague]

The nullable changes look good. I've left a few comments on SocketReader, although they mostly pertain to the original PR. TBH the benefit/risk ratio on this PR doesn't feel super high

[gbirchmeier]

I made those changes.

I have also nullable-ized and cleaned up SocketInitiatorThread, and also replicated the similar changes in that file. All tests appear passing, and I would again appreciate a pair of eyes to review these changes.

[Rob-Hague]

Looks good to me