Test binaries are relocatable

[jakirkham]

As observed in a recent issue ( rapidsai/cudf#18251 ), it is possible to create binaries that are not relocatable. Or more accurately that when trying to relocate the binaries, they can become corrupted. The issues from these corrupted binaries can be quite hairy to diagnose. To prevent this, it would be good to test that the binaries we produce are relocatable. Mostly this does happen as part of building Conda packages and testing them. However as seen in the issue above, this testing isn't always sufficient. Particularly if the relevant code is in a header-only library and the library that consumes it is in another project (like RMM, RAFT, KvikIO, etc.).

One approach to testing would be to build tests of a header-only library as a shared object that we package and then run the tests on. This would help smoke out any issues due to relocation of library. We may want to do additional auditing on this package to check for corruption, which we could also flag in CI.

[vyasr]

cf #160