Rework memory allocation failure handling in bytestring builders, add tests

[PJK]

Description

The decoder should clean up intermediate items when the callback fails. The rest of the cleanup is handled by the decoder stack unwinding

Replaces #228. In retrospect, the fix wasn't a good solution: the new behavior did fix the memory leak in the streaming decoder, but the semantics didn't make sense for standalone usage of cbor_bytestring_add_chunk and were also inconsistent with cbor_array_push.

Fixes #231

Checklist

• I have read followed CONTRIBUTING.md
  • I have added tests
  • I have updated the documentation
  • I have updated the CHANGELOG
• Are there any breaking changes?
  • If yes: I have marked them in the CHANGELOG (example)
• Does this PR introduce any platform specific code?
• Security: Does this PR potentially affect security?
• Performance: Does this PR potentially affect performance?

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 85.71429% with 1 line in your changes missing coverage. Please review.

Project coverage is 96.49%. Comparing base (118f8d3) to head (b7c04c2).
Report is 271 commits behind head on master.

Files with missing lines	Patch %	Lines
src/cbor/internal/builder_callbacks.c	85.71%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #242      +/-   ##
==========================================
+ Coverage   95.85%   96.49%   +0.64%     
==========================================
  Files          20       20              
  Lines        1544     1541       -3     
==========================================
+ Hits         1480     1487       +7     
+ Misses         64       54      -10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[PJK]

cc @James-ZHANG