Splitting from #75333

Created a task to track a suspicion that superAlignLDSGlobals is miscompiling absolute address variables at present.

Idea is to accept absolute addressed LDS on the input if and only if all variables have absolute addresses. I think this will unblock @Pierre-vh's thinlto project

Currently an absolute symbol is detected at L350 and fatal_error'ed, probably after setting alignment metadata on it which could cause trouble later.