It's true, the HTTP API documentation for the package manager went missing in one of the latest versions of the product documentation. Can you raise a Daycare ticket for it and ask to make it available again?
Regarding OSGI installer ("install" folder) vs package manager: From a reliability point of view they are the same; the only limitation is that you cannot deploy OSGI bundles via package manager, but you need to wrap them into packages. But internally both use the OSGI installer to start the bundles.
But I still prefer the package manager, because it's quite easy to make your AEM unusable with the install folder:
* copy a bundle into the install folder
* wait till it is installed
* shutdown the AEM instance
* remove that file from the install folder
* startup AEM again
(ok, this isn't a problem when you just use content packages.)
On the other hand side the package manager HTTP API is a bit problematic, because just uploading and installing packages in a row does not always work, and you need to handle the error detection, retries etc. Doable but the many trivial "automation scripts" you find in the internet only focus on the happy path. Might work in 90% of all cases.
Thanks Joerg for your feedback. I have created now a Daycare ticket for the missing documentation.
According to the "pure" OSGi Installer approach, I thought more of uploading the package or bundle directly to the "install" folder/node below /apps in the JCR repository, so not the install folder in the filesystem.
I think a main challenge here is the deployment of configurations. Depending on which configurations you upload, the Felix container restarts several or even all bundles, which causes then this Unavailability of the system for some seconds/minutes.
