Reduce the window between creating and writing to files in contributor

[ibaryshnikov]

Contributor works like this:

• calls try_lock endpoint <- the files are initialized
• download/challenge
• upload/response
• try_contribute <- the files are written to

As you can see, we have several requests before creating and using the files. And if the contributor or coordinator is stopped in between of these calls, there's a high chance to obtain incorrect state in the transcript.
What if we move the file initialization from try_lock to try_contribute? As a result, the files will be created right before being used, and it will reduce the window in which it's possible to obtain incorrect state by a lot.
If the contributor goes down, it will not affect the files at all, so we'll have to take care only about coordinator restarts.

[apruden2008]

I think this is great, as it potentially reduces the number of ways the contributor can fail.