Improve kernel ref type safety

[rekmarks]

We should use template literal types for the various refs (KRefs, VRefs, ERefs) in the kernel. We used to do this, but it was too cumbersome at an earlier, messier stage of our codebase. Not confusing these values is important, and now is the time to tighten up our use of them.

[grypez]

On hold pending rebase and due to priority. See branch for progress. Ref types factor as below.

$$ \mathrm{refs} \cong \bigg(\mathrm{kernel} \sqcup \big((\mathrm{vat} \sqcup \mathrm{remote}) \times (\mathrm{import} \sqcup \mathrm{export})\big)\bigg) \times (\mathrm{promise} \sqcup \mathrm{object}) $$

Notes:

• String validators implemented with superstructs define allow these types in rpc definitions
• KVStore made type safe using function interface, no runtime validation overhead
• Many test refs do not match their format; if no runtime validation is added, no test changes are required ;)
• Use of 'r' as prefix for both remote endpoints and remote objects seems out of place

[rekmarks]

Closed by #917 and #921