We now use interned/ref counted strings. There should be metrics to show us how effective this is. For example:
- count of strings in pool
- deduplication achieved - avg ref count relative to size of pool?
- size of strings in pool? might drive decision towards SSO
- more stats that might be useful