With SERVER-67471 we made some changes to RecordId in order to fix a regression we detected with our internal performance benchmarks. The fix made was to use unique buffers instead of shared buffers for large record ids as that made the regression go away.

In SERVER-68081 we discovered that this regression was only detected in the CPU used in our benchmarks and is not present in current CPUs.

The purpose of this ticket is to revert the unique buffer and go back to using shared buffers for large record ids.