All transport problems should be grouped to Bugsnag, or not sent

[JakeWharton]

Right now we just send the raw exception as an info level crash which causes about 20 issues to be tracked based on how it fails. There's two problems here:

• All transport-related problems are a single failure. The use of Timber in multiplatform code is a convenient but lazy way of tracking these. We should probably use something other than Timber for reporting crashes but that requires a multiplatform abstraction.
• Using a crash reporting tool for this is dumb. These aren't crashes, but they're still useful to track. A better metric would be to correlate successful fetches with failed fetches as a percentage and track that over time to ensure the number doesn't jump to 100%.

[JakeWharton]

Wow this is further complicated by the absence of IOException in the multiplatform common world.