feat: use 1 client with multiple channels for the benchmark

[mr-salty]

• factor Client creation code out to a helper
• make InsertOrUpdateExperiment use DefaultPRNG like the others
• remove superfluous mutex from UpdateDmlExperiment::RunIteration
• start to migrate from minimum/maximum_clients to _channels

Part of #1193

---

This change is 

[coryan]

Thanks, it looks good, assuming I understand the plan: after this you will change the other benchmarks and then rename the *_clients variables?

Somewhat unrelated: can you attach your results to one of the bugs so we know (in the future) why keeping a single client was enough?

Reviewed 3 of 3 files at r1.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @devbww)

[codecov]

Codecov Report

Merging #1194 into master will decrease coverage by 0.08%.
The diff coverage is 86.04%.

@@            Coverage Diff             @@
##           master    #1194      +/-   ##
==========================================
- Coverage    95.7%   95.62%   -0.09%     
==========================================
  Files         174      174              
  Lines       13601    13565      -36     
==========================================
- Hits        13017    12971      -46     
- Misses        584      594      +10

Impacted Files	Coverage Δ
...nner/benchmarks/single_row_throughput_benchmark.cc	91.93% <100%> (-0.81%)	⬇️
...ogle/cloud/spanner/benchmarks/benchmarks_config.cc	93.51% <25%> (-5.51%)	⬇️
...ud/spanner/integration_tests/client_stress_test.cc	81.57% <0%> (-1.76%)	⬇️
...loud/spanner/internal/partial_result_set_source.cc	93.24% <0%> (-1.36%)	⬇️
google/cloud/spanner/internal/spanner_stub.cc	69.23% <0%> (-1.1%)	⬇️
...on_tests/rpc_failure_threshold_integration_test.cc	85.55% <0%> (-0.16%)	⬇️
...anner/integration_tests/client_integration_test.cc	98.47% <0%> (-0.01%)	⬇️
google/cloud/spanner/keys.h	100% <0%> (ø)	⬆️
.../spanner/benchmarks/multiple_rows_cpu_benchmark.cc	92.35% <0%> (ø)	⬆️

---

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 27048c7...5e75b25. Read the comment docs.

[mr-salty]

yeah, I'm planning to do the multiple row benchmark in a different PR then make those channels.

I added a graph to https://github.com/googleapis/google-cloud-cpp-spanner/issues/1193

Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @devbww)

[coryan]

Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @devbww)