Compile profile-optimized release binaries (PGO). (Where to use the optimized binaries is a separate decision.)

The building has more steps: 1) compilation with -fprofile-generate (.gcno), 2) running tests recording the coverage data (.gcda), and 3) compilation with -fprofile-use. This needs to be done only once for the actual release. (Though: from CI stability and release process point of view it might be beneficial for the release build process to resemble the usual build process as much as possible.)

The benefit of the optimization (in any single case) depends on how representative (to the case) the tests are.

According to a very quick research (googling) this should be production-safe.