Add tests around encoding handling for various readers

[mpenkov]

smart_open supports a variety of reader protocols:

• s3
• hdfs
• webhdfs
• http
• any others?

Currently, each of those implements its own logic for handling character encoding. This logic isn't being tested anywhere.

We should add tests to test that this logic works. In the future, we may refactor such that the encoding logic gets handled in a single place for all readers, so these tests should face the public API, as opposed to something internal.