Use travis pypy3

[c24t]

Instead of pypy3.5, debugging #406.

Fixes #406.

[codecov-io]

Codecov Report

Merging #408 into master will decrease coverage by 0.09%.
The diff coverage is n/a.

@@            Coverage Diff            @@
##           master     #408     +/-   ##
=========================================
- Coverage   85.27%   85.18%   -0.1%     
=========================================
  Files          38       38             
  Lines        1956     1937     -19     
  Branches      231      230      -1     
=========================================
- Hits         1668     1650     -18     
  Misses        221      221             
+ Partials       67       66      -1

Impacted Files	Coverage Δ
...ry-sdk/src/opentelemetry/sdk/resources/__init__.py	68.18% <0%> (-2.66%)	⬇️
...ry-ext-wsgi/src/opentelemetry/ext/wsgi/__init__.py	67.59% <0%> (-0.59%)	⬇️
...opentelemetry/sdk/context/propagation/b3_format.py	86.79% <0%> (-0.49%)	⬇️
opentelemetry-sdk/src/opentelemetry/sdk/util.py	85.54% <0%> (-0.35%)	⬇️
...ts/src/opentelemetry/ext/http_requests/__init__.py	89.47% <0%> (-0.27%)	⬇️
...src/opentelemetry/ext/opentracing_shim/__init__.py	95.76% <0%> (-0.18%)	⬇️
...xt-jaeger/src/opentelemetry/ext/jaeger/__init__.py	86.39% <0%> (-0.16%)	⬇️
...app/src/opentelemetry_example_app/flask_example.py	100% <0%> (ø)	⬆️
...emetry-sdk/src/opentelemetry/sdk/trace/__init__.py	90.67% <0%> (+0.3%)	⬆️

---

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 ad7a809...d3781d9. Read the comment docs.

[c24t]

Looking back at #95, it's not clear why we're using pypy3.5 here instead of pypy, which the travis docs recommend.

Not as satisfying as solving the actual underlying travis issue, but we can unblock builds with this PR now and revisit if we have a reason to target pypy3.5 specifically in the future.

[codeboten]

Nice, thanks for fixing this

[c24t]

Entering god mode to merge and unblock builds.