Improve drift handling w.r.t. throttling

[matthchr]

For large numbers of nodes with a lot of churn ongoing, the drift controller will call IsDrifted quite frequently.

IsDrifted will call GetNIC. This can trigger throttling when it's done frequently.

There are also calls in isImageVersionDrifted to GET VM, which could also trigger throttling although based on some testing by partners, it doesn't hit throttling as easily as GET NIC does.

In both cases, we can probably do away with the call to Azure by:

1. Caching the data we need either in-memory or in an annotation/hash on the nodeClaim.
2. Only doing a real GET to Azure if the cached data suggests we need to.

[tallaxes]

+1 to avoiding calls to Azure, #486 (comment)