These random failures happen because FD number allocation
is not predictable when multiple threads are running (since
MJIT thread is opening/closing files all the time).

Real-world code practically never relies on predictable FD
number allocation (because much real-world code is
multi-threaded); so it's highly unlikely there'll be any
breakage to the user.

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@66159 b2dd03c8-39d4-4d8f-98ff-823fe69b080e