liveslots should free Remotables upon dispatch.dropExport

[warner]

What is the Problem Being Solved?

Once the VOM is holding on to Remotables wherever necessary (#3132), it is safe to have liveslots process dispatch.dropExport properly. This is the kernel's signal that neither the kernel nor any other vat can still reach the Remotable.

Upon receiving dropExport, liveslots should remove the Remotable from the (strong) exportedRemotables Set. This removes liveslots' last strong reference to the Remotable. If userspace is not keeping it alive in some other way (and the VOM isn't keeping it alive), the Remotable will eventually be released, along with any data it was holding in place. We expect this will free the bulk of the objects currently held by our vats.

(Note that the kernel won't actually call dropExport until #2946 is complete)

This is a component of #3106

Test Plan

A unit test (probably test-liveslots.js will build a liveslots instance without a kernel, and a simple vat. We'll use vatPowers to smuggle in WeakRef, so the vat code can sense if/when a Remotable goes away. Then the unit test will trigger dropExport and force a GC pass, then query the vat code to check that the WeakRef is empty.