Ensure a deterministic diff result

[iwahbe]

Fixes #876

This prevented pulumi/pulumi-gcp#1785 from merging.

[VenelinMartinov]

"But people might start depending on the order!"
Thanks for fixing - I've had to escape a few diffs in GRPC tests with "*" as well

[VenelinMartinov]

Actually could this have any serious performance implications?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 59.21%. Comparing base (f6013b6) to head (2ac2344).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1737      +/-   ##
==========================================
- Coverage   59.74%   59.21%   -0.53%     
==========================================
  Files         301      308       +7     
  Lines       42034    42416     +382     
==========================================
+ Hits        25112    25116       +4     
- Misses      15488    15866     +378     
  Partials     1434     1434              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[iwahbe]

Actually could this have any serious performance implications?

I'd be very surprised if the number of properties being sorted were large enough to have performance implications. Especially compared to the overhead of serialization, IPC, and network requests.

[t0yv0]

Probably fine, can't think of why the ordering would matter in prod, but might trip up some overfitting tests. A bit odd to change prod code to make testing easier. But I can see the motivation.