Error: move << operator from .hh to .cc file

[scpeters]

🦟 Bug fix

Fixes seg-fault of test in ign-physics when printing sdf::Error messages.

Summary

An ign-physics test was observed to seg-fault in gazebosim/gz-physics#261 when printing sdf::Error messages (see gazebosim/gz-physics#261 (comment)). From that comment:

I can now reproduce it with sdformat build from source. The trick is to build it with g++ 7.5 as it's done by the nightly debbuilder. My new hypothesis is that sdf::operator<<(std::ostream, const std::Error&) is inlined in ign-physics code because it's defined in sdf/Error.hh. And since ign-physics is compiled with g++ 8, there might be some weird ABI issues between symbols accessed by the inlined code and the ones available in libsdformat.so.

I see two options:

1. Move sdf::operator<<(std::ostream, const std::Error&) to Error.cc so it won't get inlined.
2. Use g++8 in the sdformat nightly debbuilder job.

This takes the second approach. I'm interested to see if the ABI checker is ok with this.

Checklist

• Signed all commits for DCO
• Added tests
• Updated documentation (as needed)
• Updated migration guide (as needed)
• codecheck passed (See contributing)
• All tests passed (See test coverage)
• While waiting for a review on your PR, please help review another open pull request to support the maintainers

Note to maintainers: Remember to use Squash-Merge

[scpeters]

CI is not happy; maybe @azeey has a better idea how to implement this?

[azeey]

Fixed compiler error and visibility issues in 1432e97

[scpeters]

Fixed compiler error and visibility issues in 1432e97

thanks!

[scpeters]

@azeey I'll let you merge if you approve

[scpeters]

backport to sdf9 in #625