Races in the PackagedFunctionRunner

[negz]

What happened?

While fixing some racy locking code in #5651 I realized I used a similar anti-pattern in the composition function PackagedFunctionRunner. There's a couple of places where that code:

1. Takes the read lock, reads some state, and releases the read lock
2. Computes something based on the read state
3. Takes the write lock, updates the state, and releases the write lock

The risk is that another Goroutine takes the write lock between when we release our read lock and take our write lock, thus invalidating the computed state we write back.

How can we reproduce it?

I guess run a lot of functions.

What environment did it happen in?

Crossplane version: