fix: fix flaky BulkWriter test

[thebrianchen]

Fixes #1097.

I couldn't reproduce this locally, but my hypothesis is that the test had floating promises that resulted in race conditions. If the second flush() call was made before the first flush() call sent its request, the rate limit threshold would not have have been reached for the second flush(), which meant the timeout handler that ended the test would never be invoked.

By waiting for promises to complete instead of using done(), the test should now execute in the correct order each time.

[codecov]

Codecov Report

Merging #1115 into master will increase coverage by 1.20%.
The diff coverage is 98.66%.

@@            Coverage Diff             @@
##           master    #1115      +/-   ##
==========================================
+ Coverage   97.40%   98.61%   +1.20%     
==========================================
  Files          27       28       +1     
  Lines       16978    18301    +1323     
  Branches     1271     1405     +134     
==========================================
+ Hits        16538    18047    +1509     
+ Misses        436      251     -185     
+ Partials        4        3       -1     

Impacted Files	Coverage Δ
dev/src/field-value.ts	97.06% <92.85%> (+0.01%)	⬆️
dev/src/types.ts	99.39% <93.33%> (-0.61%)	⬇️
dev/src/bundle.ts	93.92% <93.92%> (ø)
dev/src/index.ts	98.17% <95.51%> (+0.42%)	⬆️
dev/src/v1/firestore_client.ts	97.86% <98.85%> (+3.05%)	⬆️
dev/src/document.ts	98.77% <98.92%> (+0.15%)	⬆️
dev/src/v1beta1/firestore_client.ts	97.38% <99.02%> (+2.89%)	⬆️
dev/src/reference.ts	99.88% <99.54%> (+0.07%)	⬆️
dev/src/backoff.ts	100.00% <100.00%> (ø)
dev/src/bulk-writer.ts	98.85% <100.00%> (+0.36%)	⬆️
... and 29 more

---

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 8ffe256...4d063c1. Read the comment docs.

[schmidt-sebastian]

Fix LGTM. I think we should fix in master since #1097 is against master.