Improve memory efficiency of EntityCache

[lutter]

This PR makes the EntityCache more memory efficient, meaning the same amount of memory will be able to store more entities with two changes:

• Use Word (Box<str>) for entity types and ids
• Remove the subgraph id from EntityKey since that adds an overhead of 70 bytes for every cache entry

[leoyvens]

Code lgtm, good thing we didn't have EntityKey in the actual ProofOfIndexingEvent. The change is very mechanical but it touches many critical pieces of the code, so might be worth an integration cluster run.

[lutter]

I ran this in the integration cluster for > 24hrs without any incident. Merging it therefore