Fixes issue #308.

Non-fitting text is truncated by simply cutting it off.

win_textwidth() has been replaced by a simple macro wrapper around
win_draw_text() with a maximum width of zero, which results in a dry-run that
still calculates the number of pixels used by the text, much like snprintf(3)
with a buffer length of zero.

Code under a different license should be kept in a separate file. This
implemention is a single header file with ~65 lines, so it better fits this
requirement.

Fixes #276

Instead of rendering the entire filename at once, Xft will let us do it
character by character. This will allow sxiv to query fontconfig for
a font that can provide any missing codepoints, if needed.

A known issue of this patch is that the "..." dots rendering will not
work properly for very long multibyte filenames. That is because we
cannot easily predict the final width of the rendered filename before
drawing it. I couldn't figure out a clean way to deal with this, so I
ended up just truncating the offending filenames.

None of the mouse mappings uses a keyboard modifier, making it possible to
access the most basic features by only using the mouse.

Next/previous image with left button depending on cursor position, middle
button for dragging, right button for switching to thumnail mode and wheel for
zooming.

Users can keep the old behaviour by simply not adapting the changes to the
buttons array in config.def.h to their config.h file.

This makes mouse panning more direct and faster.

- Functions warn() and die() replaced by GNU-like error(3) function
- Register cleanup() with atexit(3)
- Functions called by cleanup() are marked with CLEANUP and are not allowed to
  call exit(3)

Old snprintf calls could have overflowed the buffers.

- Buffer for window content is bigger than the window, minimizes artifacts
  when window is resized
- Back to using XSetWindowBackgroundPixmap() instead of XCopyArea(),
  no need to handle exposure events; X server can show gray background directly
  after resize event before sxiv redraws the window contents

Without this some window managers (e.g. fvwm) will ignore the initial
window position and place it according to it's own rules.

- Draw onto pixmap as before, but use the same size for the pixmap as
  for the window, allocate new pixmap after configure requests
- Use XCopyArea() instead of XSetWindowBackgroundPixmap(), which now
  requires handling of Expose events