Improved error-handling in S3 completion callback

[graebm]

Issue:
A user experienced random downloads hanging. I looked at their code and saw that an exception in their completion callback could lead to a hang, since later code was never executed. Then I looked at our code and saw a theoretical place where the same kinda thing could happen, an error could prevent us invoking the completion callback.

Description of changes:

• Always invoke the completion callback, even if we failed to turn the error_response_headers into PyObjects. These headers are optional anyway.
• Use nice new error code AWS_ERROR_FILE_WRITE_FAILURE if file write fails.
• Raise error if file close fails, because that probably means it failed to flush the buffers to disk.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.