Use patched filelock with CLOEXEC

[snoyberg]

Specifically, we're adding in this commit:

snoyberg/filelock@4f08049

Without this change, very often on large builds we end up in a situation
where:

• Multiple threads are working in parallel
• Thread A needs to work with the database
• Thread A takes a file lock on the database
• Thread B forks a child process, which inherits the database lock
• Thread A releases the file lock
• Thread C attempts to take the file lock, but is blocked by B's child

As a demonstration, lsof showed me this on my system:

COMMAND     PID    USER   FD   TYPE DEVICE SIZE/OFF       NODE NAME
stack     20216 michael   12w   REG    1,4        0 8613331527 stack.sqlite3.pantry-write-lock
Cabal-sim 22736 michael   12w   REG    1,4        0 8613331527 stack.sqlite3.pantry-write-lock
ghc       22800 michael   12w   REG    1,4        0 8613331527 stack.sqlite3.pantry-write-lock

With this change in place, I witnessed 0 cases of the file lock not
acquired message from Stack on a large build.