Mitigate race conditions in Adapters

[dbemiller]

In this project's short lifetime, we've already started to see how easy it is to introduce race conditions (#122, #140).

This is a problem, because:

1. The project has no automated way to catch them. Only code review & comments.
2. Adapters are run in their own goroutine, so segfaults crash the entire process.

Programmers make mistakes, and anything which isn't caught automatically will happen in the future. As this project becomes more active, this will become a bigger issue.

There are a few options for how we can mitigate these issues.

1. Copy every object we send into the Adapters' goroutines, so that data races aren't possible.
2. Add "good adapter citizen" tests which run on all adapters, and make sure they don't mutate shared state.
3. Change the Types we send into Adapter.Call() so that adapters can't mutate things they're not allowed to. For example, make the PBSRequest state private and expose methods which only return immutable data.

Anyone have preferences, or other ideas?

[mkendall07]

I like option 3. Although 2 is probably also a good idea. 1 seems a bit heavy handed. That's all I got. no new ideas.

[dbemiller]

There's also some overlap here with #150.

If the PBS external API changes to accept openRTB requests, then the pbsrequest type will go away, and the adapter API will have to change.

[dbemiller]

#356 should (I hope) make Travis catch many of these. If we still find that we're running into problems, we can look into ways to improve the design farther.