This patch changes the following slots of `struct window'
from Lisp_Object to:

   use_time -> int
   sequence_number -> int
   last_modified -> EMACS_INT
   last_overlay_modified -> EMACS_INT
   last_point -> ptrdiff_t
   window_end_pos -> ptrdiff_t
   window_end_vpos -> ptrdiff_t

Notes:

1. IIUC, `window_initialized' is redundant and obsolete.

2. Fields hscroll and min_hscroll are good candidates for conversion
    too, but this requires some redesign of `struct saved_window', most
    probably to yet another pseudovector.

Special notes about hard stuff in xdisp.c:

1. Field window_end_valid is used in a tricky way which isn't quite
    clear for me :-), so I would like to not touch it at this time
    (hopefully I will get back to it later).

2. Since last_point is an integer, this check is no longer valid:

    /* Right after splitting windows, last_point may be nil.  */
    && INTEGERP (w->last_point);

    My tests never shows zero here, so I'm installing assertion instead.

3. The check:

    && INTEGERP (w->window_end_vpos)
    && XFASTINT (w->window_end_vpos) < w->current_matrix->nrows

    is transformed to:

    && w->window_end_vpos > 0
    && w->window_end_vpos < w->current_matrix->nrows

    I never get w->window_end_vpos >= w->current_matrix->nrows, but
    run into w->window_end_vpos == 0 several times. At this moment, I have
    no ideas whether Gerd was correct about fixing window.c, so this
    needs more detailed investigations.

Dmitry