Address misreported value on exemplars for counters.

[jdmarshall]

After one AI user filed a PR to fix this the wrong way, I finally went tracing through the code to understand what the heck is wrong as implemented. The real issue turned out to be pretty dumb: mislabeled variables resulting in mischaracterization of the data it represents.

We're calling the increment step 'value' but we use the 'value' to call setDelta in incWithoutExemplars() and delegating to it from incWithExemplars() but then using the passed-in delta as a 'value' again when updating the 'value' field in the exemplar, which is not the same meaning of 'value'.

Within minutes of updating the ticket to set the labels and the problem, a completely separate user filed a PR. Maybe I should not have set helpWanted when a PR was already open. However, both PRs had an identical new unit test, and both of them were identically asserting the wrong thing, and solving the issue the wrong way.

At this point I had invested more time in both PRs than the authors had. The entire problem was front-of mind, and the existing test is wrong but written in a way where either interpretation would be correct. Clearly they are not.

Fixes #621

[jdmarshall]

I initially had planned to rename the variables to make the distinction clear, but unfortunately the 'value' is exposed as part of the public API and I don't want to break existing code. Even though technically any existing code is already broken.

[jdmarshall]

@zbjornson This is pretty much the last PR I really want in 0.16 if you could take a look.

[zbjornson]

Makes sense.