Refactor transcode loop context and improve tests stability

**Is your feature request related to a problem? Please describe.**
We currently have little control/visibility on the transcode loop process, so our tests around the loop rely on waiting for the timeout. We want to refactor the context such that tests can inject their own stub. With a stub context we can write more reliable and readable tests.

**Describe the solution you'd like**
See @j0sh's suggestion [here](https://github.com/livepeer/go-livepeer/pull/657#discussion_r244917469). In short we want to refactor the creation of the transcode loop context into a function we can stub out in tests.

**Describe alternatives you've considered**
Currently tests rely on a combination of waiting for the timeout and checking for some post-timeout side effects, but we recognize the current approach as brittle and inflexible.

**Additional context**
None.

Start by locating the transcode loop context creation and read the suggestion in pull request #657, discussion r244917469. Refactor that creation so tests can provide a stub context, then update the loop tests to avoid waiting on timeouts and verify the expected post-timeout behavior.

Written by the indexing model from the issue text.