Fix performance regression on Cloud Pub/Sub publish

[kamalaboulhosn]

With the refactor to support ordering keys, we moved away from publishing on an executor in the unordered case. This causes too many threads to build up and we end up with a lot of timeouts for publish. This PR restores to the way the code was prior to this refactor.

[hannahrogers-google]

LGTM

[codecov]

Codecov Report

Merging #6076 into master will increase coverage by 0.31%.
The diff coverage is 100%.

@@ Coverage Diff @@## master #6076 +/- ##
============================================
+ Coverage 47.11% 47.43% +0.31% 
Complexity 27195 27195 ============================================
Files 2522 2522 Lines 274268 274271 +3 Branches 31326 31350 +24 ============================================
+ Hits 129235 130102 +867 + Misses 134602 134559 -43 + Partials 10431 9610 -821

Impacted Files	Coverage Δ	Complexity Δ
...ain/java/com/google/cloud/pubsub/v1/Publisher.java	89.11% <100%> (+0.11%)	40 <0> (ø)	⬇️
...m/google/cloud/bigquery/QueryJobConfiguration.java	75.15% <0%> (+0.31%)	50% <0%> (ø)	⬇️
...om/google/cloud/bigquery/LoadJobConfiguration.java	88.73% <0%> (+0.45%)	46% <0%> (ø)	⬇️
...ogle/cloud/bigquery/WriteChannelConfiguration.java	86.12% <0%> (+0.47%)	44% <0%> (ø)	⬇️
...n/java/com/google/cloud/bigquery/BigQueryImpl.java	66.78% <0%> (+0.52%)	57% <0%> (ø)	⬇️
...java/com/google/cloud/firestore/FirestoreImpl.java	83.42% <0%> (+0.57%)	22% <0%> (ø)	⬇️
...main/java/com/google/cloud/storage/BucketInfo.java	84.86% <0%> (+0.57%)	81% <0%> (ø)	⬇️
...ud/spanner/jdbc/AbstractJdbcPreparedStatement.java	90.85% <0%> (+0.6%)	52% <0%> (ø)	⬇️
.../com/google/cloud/compute/deprecated/DiskInfo.java	96.5% <0%> (+0.69%)	29% <0%> (ø)	⬇️
.../cloud/scheduler/v1beta1/CloudSchedulerClient.java	55% <0%> (+1%)	35% <0%> (ø)	⬇️
... and 139 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 7441cce...173adc5. Read the comment docs.