Implement ENOSPC fault injection for WiredTiger

We should provide infrastructure that can be used to inject ENOSPC errors in WiredTiger tests.  

WT-4065 discusses a few approaches for doing this.  

A reasonable approach would be to build on the existing fail_fs file system extension.  A simple implementation would simply return ENOSPC to all space-allocating operations after a trigger point (e.g., after X MB of data has been written).  A more sophisticated approach would track the total size of all files and return ENOSPC to operations that would exceed a trigger threshold.  This would mimic the observed behavior where a fallocate request can fail to preallocate a lot of space to a log file, but subsequent (smaller) allocating writes succeed.