Allows arbitrary sized windows on Darwin... the OS is still scaling
the window contents though rather than allowing us to dynamically
update them.

Whoops.

This should normalize everything, with the possible exception of
"zoom" (the thing other OSs call maximized).  Ideally, we wouldn't
allow window size changes when "zoomed" if we could detect the state,
but it doesn't look like we can.

Fixes sourceforge issue 126 (finally!)

Apparently, macOS "maximized" is the same as "fullscreen" in SDL.

The issue was that the various semaphores and mutexes weren't being
initialized at all when sdl_initconio() wasn't called... so starting
the event thread and telling it to stop would access uncreated
semaphores and mutexes.

Split out the init, and be sure to call it before exiting.

1) Use static SDL since it's required by both conio and XPDev
2) Since the SDL thread will always run, any attempt to shut down
   the SDL thread calls exit(0).  Not exactly what we want to
   return, but at least this makes it not hang waiting for the
   impossible.

Should fix js test suite on darwin.

Translation from mouse coordinates to screen coordinates had issues
with internal vs. external scaling and for when there are gutters
(ie: black bars) on the edges.

This change snaps all gutter clicks to the nearest position inside
the window instead of discarding them, and ensures the calculations
are correct regardless of scaling mode.

Win32GDI mode was not tested, the only change there was to stop
ignoring mouse positions in the gutters.

do math to get there.

With 0xE0 being used for ciolib "super-extended scancodes", a literal
0xe0 can't pass through the input path.  This is an issue in CP866 (р)
and KOI8-U (Ю) as well as CP437 α.

Should fix SyncTERM SF bug 123.

For example, if it was set to 0.4, the initial scaling would be set
to take at most 40% of the screen height or width (whichever is
smaller) with a minimum of 1.0

Add some malloc() return checks, comment fall-throughs, and resolve
a sleep while holding lock.

None of these should actually matter, so hopefully I didn't add a
new bug.

When using the marco WM, and resizing using ALT-Right-drag, calls to
SDL_SetWindowMinimumSize() result in the top-left corner of the
window moving up and to the left (appears to be by the border size).

To prevent this from being a maddening issue under marco, ensure
we only call SDL_SetWindowMinimumSize() once when the minimum size
changes on the window.

Fixes SF ticket 115, thanks Ragnarok!

While we're hear, ensure the minimium maximum window size holds
the original sized window... we're not interested in downscaling.

sdl_getscaling() does not require vstatlock.  Move it out.

Currently useful only in GDI and SDL modes.  The external scaling
method is likely lower CPU utilization and may or may not look
better.

All bitmap drivers now support arbitrary scaling, and the scaling
factor is a double, allowing arbitrary window scaling in all bitmap
modes (making nelgin happy).

While we're here, fix bugs in horizontal interpolation and X window
resizeing.

For graphical modes, this requests that ciolib be initialized in
the specified mode.  For text-based modes (ANSI, Curses, Win32 Console),
this is not used and "current" is used instead.

Pretty sure this is what intptr_t was created for.

This is now simple to implement.

Also, add ciolib_initial_scaling so the initial window can be created
with the appropriate scaling already applied.

We still need a way to specifiy the initial text mode, but these
globals are not the right way.

Move snap logic into bitmap
Move new mode window size logic into bitmap

This caused double-keys on those when numlock was off.  Only the
number keys and period need special NumLock handling.

Fixes bug 108 on SourceForge, thanks nelgin!

Add setwinsize and setwinpos
Move blit from rect into gdi_thread
Add beep support
Add (non-functional) settitle support
Support ciolib scaling in addition to GDI scaling
Add window-size snapping
Support -iG in SyncTERM

GDI is still the second-least preferred mode, so auto mode will only
use it if the windows console somehow fails.  But at least people
can play with it.

Fixes issue on Linux where the window size is constantly updated,
and window rabidly steals focus.

This lock has devolved into a wrapper lock for vstatlock.

Use pthread_once() for init_mouse() instead of a global
Fix some lock issues in cb_drawrect()

When fast scrolling, memmove() takes most of the CPU by far... it
is likely actually worth having each line be a buffer of its own
and just moving the pointers around.  To help with this optimization
in the future, move the screen memmove() bit into a separate function.

Fix a couple warnings while we're here.

This should almost completely remove vstatlock contention in SyncTERM,
which should bring SDL output performance close to X11 performance.

This makes cvstat work as intended, so we no longer need to lock
vstatlock when accessing it since it's only accessed from the
video event thread.

There's still an irritating dependency on vstat for ALT-Enter handling.