Optimize indexing for file-system-centric deployments

In OAK-5192 we started discussing potential short / mid term solutions for reducing the IO load of Oak Lucene index on the repository, however it would be good to discuss also if we can sketch out a longer term solution (probably for segment tar only) that would avoid writing some binaries in the repo in first place relying more on the file system, but without incurring in the old JR2 problems (e.g. index corruptions).