Fix CI benchmark

[marcalff]

Fixes #1800

Changes

Fix OpenTelemetry-cpp benchmarks / Run OpenTelemetry-cpp benchmarks in CI

The ubuntu-latest upgrade from 20.04 to 22.04 broke benchmarks.
Restoring ubuntu-20.04

• CHANGELOG.md updated for non-trivial changes
• Unit tests have been added
• Changes in public API reviewed

[marcalff]

@open-telemetry/cpp-approvers

This is a blind fix, it can not be tested on PR push,
because the benchmark workflow runs on main only.

Most likely it will fix the break,
and if not it will be trivial to revert anyway.

[codecov]

Codecov Report

Merging #1799 (2f76ff6) into main (6ba1f73) will decrease coverage by 0.08%.
The diff coverage is n/a.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1799      +/-   ##
==========================================
- Coverage   85.79%   85.71%   -0.07%     
==========================================
  Files         171      171              
  Lines        5240     5240              
==========================================
- Hits         4495     4491       -4     
- Misses        745      749       +4     

Impacted Files	Coverage Δ
ext/src/http/client/curl/http_client_curl.cc	80.31% <0.00%> (-1.13%)	⬇️
sdk/src/trace/batch_span_processor.cc	90.63% <0.00%> (-0.78%)	⬇️

[ThomsonTan]

@open-telemetry/cpp-approvers

This is a blind fix, it can not be tested on PR push, because the benchmark workflow runs on main only.

Most likely it will fix the break, and if not it will be trivial to revert anyway.

Can we run benchmark on PR build too?

[lalitb]

Can we run benchmark on PR build too?

It was removed from PR build purposefully, in order to not have heavy load on servers when benchmark test is running.

[esigo]

@open-telemetry/cpp-approvers
This is a blind fix, it can not be tested on PR push, because the benchmark workflow runs on main only.
Most likely it will fix the break, and if not it will be trivial to revert anyway.

Can we run benchmark on PR build too?

That would pollute our benchmark graphs.

[marcalff]

To clarify, this fixes:
OpenTelemetry-cpp benchmarks / Run OpenTelemetry-cpp benchmarks
not to be confused with
CI / Benchmark