Convert tests to from :sdk:metrics-testing to :sdk:testing

[jack-berg]

Wanted to increase confidence that :sdk:testing has all the tools needed for making metrics assertions after realizing that internally we're still using :sdk:metrics-testing everywhere. So I switched all our internal usage to :sdk:testing.

For the most part, the tools in :sdk:testing suffice. In rare cases like asserting the size of point data it was possible but more awkward. In cases where we test directly test low level features like exemplar we don't have the accessors to make assertions directly (everything flows through assertThat(MetricData)).

I found that HistogramPointAssert was missing methods to assert exemplars, so I added those.

I deleted everything in :sdk:metrics-testing that was no longer being used. What's left are some utilities for asserting exemplars directly, and for asserting exponential histograms.

[codecov]

Codecov Report

Merging #4444 (5e138de) into main (89c6323) will increase coverage by 0.06%.
The diff coverage is 100.00%.

@@             Coverage Diff              @@
##               main    #4444      +/-   ##
============================================
+ Coverage     89.95%   90.01%   +0.06%     
- Complexity     4957     4970      +13     
============================================
  Files           569      570       +1     
  Lines         15358    15384      +26     
  Branches       1470     1472       +2     
============================================
+ Hits          13815    13848      +33     
+ Misses         1073     1067       -6     
+ Partials        470      469       -1     

Impacted Files	Coverage Δ
...etry/sdk/testing/assertj/HistogramPointAssert.java	100.00% <100.00%> (+14.81%)	⬆️
...telemetry/sdk/testing/assertj/HistogramAssert.java	100.00% <0.00%> (ø)
...ava/io/opentelemetry/sdk/internal/RateLimiter.java	100.00% <0.00%> (+5.88%)	⬆️
...telemetry/sdk/testing/assertj/LongPointAssert.java	100.00% <0.00%> (+15.38%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 89c6323...5e138de. Read the comment docs.

[jsuereth]

My only comment here is that some of the tests used satisifiesExactly and some used anySatisfy for points. It seems the new testing library either doesn't support that use case, or we need some other assertions in addition here.

Specifically, some of the "satisfiesExactly" are in place to prevent certain bad behaviors, while the "anySatisfy" are because testing exact behavior is hard.

[anuraaga]

We can add assertions that seem useful indeed. I wouldn't expect to need satisfiesExactly - our data model / OTLP doesn't specify anything about the order of points within a metric AFAIK, so presumably we wouldn't want assertions with strict ordering. anySatisfy on the flipside seems very imprecise as it only matches one point so also not sure about adding this - it's always possible to break out of fluent assertions to use normal assertj matching, and if we see a lot of that in the wild then we'd want to add it.

[anuraaga]

Thanks @jack-berg completely missed doing this