[SEMANTIC CONVENTIONS] Upgrade to 1.21.0

[marcalff]

Fixes #2138

Changes

Upgrade:

• Semantic conventions to 1.21.0
• Build tools to 0.19.0

Adjust code per the following semantic convention changes:

• BREAKING: Replace net.peer.*/net.host.* with client.*/server.* (and source.*/destination.*) opentelemetry-specification#3402
• BREAKING: Introduce common url.* attributes, and improve use of namespacing under http.* opentelemetry-specification#3355

For significant contributions please make sure you have completed the following items:

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

[codecov]

Codecov Report

Merging #2248 (ff2e9da) into main (0c5f90d) will increase coverage by 0.03%.
The diff coverage is n/a.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2248      +/-   ##
==========================================
+ Coverage   87.30%   87.32%   +0.03%     
==========================================
  Files         169      169              
  Lines        4897     4897              
==========================================
+ Hits         4275     4276       +1     
+ Misses        622      621       -1     

see 1 file with indirect coverage changes

[lalitb]

LGTM Thanks for the upgrade. Not related to this PR, but the trace and resource header files are quite big now. Also, someone interested in only DB-specific instrumentation will have HTTP-related static variables getting created and never used.
You know this part better, but is it feasible to generate domain-specific header files, say trace/semantic_convention_db.h, trace/semantic_convention_http.h etc? We can still have the existing trace/semantic_convention.h which includes all these domain-specific header files. If feasible, we can have an issue created for someone to pick.

[marcalff]

LGTM Thanks for the upgrade. Not related to this PR, but the trace and resource header files are quite big now. Also, someone interested in only DB-specific instrumentation will have HTTP-related static variables getting created and never used. You know this part better, but is it feasible to generate domain-specific header files, say trace/semantic_convention_db.h, trace/semantic_convention_http.h etc? We can still have the existing trace/semantic_convention.h which includes all these domain-specific header files. If feasible, we can have an issue created for someone to pick.

This can be investigated.

The risk I see is that the build scripts and the header file layout produced will be more coupled to the internal code organization in the semantic-convention/model directory:

• If a new yaml file is added, we may not notice it, forgetting to generate semconv for it
• If a yaml file is split, moved, merged, renamed, etc, the header files generated for it will be affected, breaking C++ users who include this header file alone.

I would rather wait for a couple of semantic-conventions releases first, to make sure the code layout itself is stable, before we can consider this.