Previously we used the "script" command-line utility to capture output from
programs that expected to be run in a terminal.  
Now, we use openpty and other C function (as wrapped by the unix library) to
run them directly from Haskell.  

As a related change, we now return the individual chunks of bytes that were
output from the process that we're testing.  This lets us test its behavior
more specifically, in particular in the "file style" tests with an incomplete
byte sequence at EOF.

This fixes the build on ghc>=7.6.1, which no longer provides the
extensible-exceptions package by default.  (Haskeline's own
dependency on extensible-exceptions was removed in haskeline-0.7.0.0.)

Patch from Paolo Capriotti.

This removes the dependency on the utf8-string package (except for ghc<7.4.1).
A few minor encoding/decoding corner cases were also fixed.

A couple cases which are known to have triggered this are:
- Entering a long (>2^15) input line
- Tab-completing a folder containing many files, with completionPaging:False

Here "large" means an area of >2^15 cells, e.g., 200x200.
Original patch by fryguybob@gmail.com.

Although GHC provides similar functionality by default, it will
terminate the program the second time that ctrl-c is pressed.  In contrast,
wrapInterrupt will throw an exception on every ctrl-c, which is better
behavior for the types of programs that use haskeline.

Also remove "import Prelude hiding (catch)" for ghc>=7.5, since
Prelude.catch will be removed from base>=4.6.

This patch switches our MonadException implementation to an approach
based on the monad-control package.

The main immediate benefit is that we no longer get warnings from the deprecated
block/unblock functions, since we now call Control.Exception.{bracket,finally,etc}
directly instead of reimplementing them from more primitive operations.

This prevents warnings about the stdcall calling convention on Win64.
This patch was submitted by Ian Lynagh.

Haddock displays those instances (which are in actually hidden to the user).
This makes it seem like InputT implements MonadState/Reader from the mtl
package.

This was also  preventing Haddock from parsing the code.

This patch is a modified version of one submitted 
by Paul van der Walt.
It adds a "haskeline_" prefix to all of Haskeline's
C symbol names.  This avoids clashing symbol names with other
packages that implement the same C functions.  
(In particular, it prevents a clash with vty on "mk_wcswidth".)