Add 3 helper APIs on top of MITM callbacks

[kegsay]

Part of #68

These helper APIs all have proven needs in existing tests, and this commit converts at least 1 test for each type of helper API to demonstrate how they can be used.

• callback.SendError: the simplest of the new APIs. This sends an HTTP error instead of the response a configurable number of times, after which point it passes the response through. Useful for testing that clients retry endpoints.
• callback.PassiveChannel: this sniffs callback data and sends it on a buffered (non-blocking) or unbuffered (blocking) channel. Useful for testing that the client sent a request.
• callback.ActiveChannel: this is a passive channel but it also allows responses to be sent back for each intercepted callback. As such, it is always an unbuffered (blocking) channel. Useful for performing an action when the test sees a certain callback.

Channels exist to reduce the amount of concurrent code that needs to be written for each test. It does this by:

• providing synchronisation points,
• with built-in timeouts,
• and handles common channel gotchas (e.g sending on a closed channel panics)

Tests can just using normal Go chan, but then they will need to reimplement these things for each test.

A good example of how this can simplify code is to look at how this PR changes testUnprocessedToDeviceMessagesArentLostOnRestartJS which closes the client when it detects a /sync response. Previously, we needed 2 additional goroutines and multiple waiters to synchronise between them. Using a single ActiveChannel, we only need 1 additional goroutine (to call StartSyncing() as that blocks) and that's it. The channel manages the synchronisation between the callback goroutine and the test goroutine, meaning the test goroutine itself can manage checking the callback data, whereas this had to be done in the callback goroutine before. It is always preferable for the main test goroutine (the goroutine running TestWhatever) to deal with data, as that means the test needs no additional synchronisation primitives to control access to that data.

[andybalaam]

Looks good, and makes sense as far as I understand it. Couple of questions and tiny suggestions.