The branch coverage reported by lcov is weird. The code before and after has the same Godbolt assembler, but the code with the lambda has a better branch coverage.
Beside the fix discussed in #961, we also had to re-adjust a test case. It seems that it was failing before, and I "fixed" it to work with the broken implementation...
* Rename 'develop' folder to 'include/nlohmann'
* Rename 'src' folder to 'single_include/nlohmann'
* Use <nlohmann/*> headers in sources and tests
* Change amalgamate config file