This should allow fall-through to SDL and win32 console if possible.

jsexec is an extra weird case though, so don't touch that one for
now... I assume it'll just break as-is, but maybe it'll work...
Certainly gdi_ciolibinit() needs better error detection.

Full-screen console doesn't scale anymore on modern Windows, so this
flag is worse than useless.

Also, fix up window behaviour... allow dragging it larger and smaller,
disable maximize, and don't allow snapping bigger or smaller than
the user area.

Finally, when r2y tables aren't installed, set the aspect ratio to
xsize x ysize so all the aspect "stuff" works properly.

With this, GDI is now the default output mode for Windows.  Auto
order is now:

X11 -> GDI -> SDL -> Win32 Console or Curses -> ANSI

Now that GDI is self-contained, does not require SDL, and runs in
a window, compile as a Windows application to prevent a console
window from opening then closing at startup.  Win32 console mode
now has to rely on AllocConsole(), which opens a new console
window.  This is likely the best choice anyway since ciolib
absolutely destroys your console settings, so that console is even
less useful than the normal windows console after a utility runs.

Maybe for 2.0 I'll do a Wayland and Core Graphics driver too. :D

Also, fix up window behaviour... allow dragging it larger and smaller,
disable maximize, and don't allow snapping bigger or smaller than
the user area.

Finally, when r2y tables aren't installed, set the aspect ratio to
xsize x ysize so all the aspect "stuff" works properly.

With this, GDI is now the default output mode for Windows.  Auto
order is now:

X11 -> GDI -> SDL -> Win32 Console or Curses -> ANSI

Maybe for 2.0 I'll do a Wayland and Core Graphics driver too. :D

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

We don't just want the kbwait() problem back, only wait after WIN_DYN
if there was no input.

Put that in your pipeline and smoke it.

Not a huge deal, but there it is (or was).

Instead, have the r2yptr and y2rptr pointers that need to be set
to those tables in order for interpolation to work.  Not setting one
locks ciolib into "LCD" modes (everything uses square pixels), and
XBR scaling is disabled.

The CIOLIB_INTERPOLATE_OBJS make variable contains the path to the
object file that contains those tables, so it's just a matter of
some code and an extra (128MB) object to link against.

Fixes high CPU utilization at menus.

This increases the binary size by 128MB, but allows much faster
startup, and should actually use less memory since the VM can page
out (or just not page in) the parts of the table it doesn't use,
and it can share the pages between different instances.

Use a static char string list for the main menu options. These can't be const
becaue uifc.list() *can* modify them (!) - if they're too long.

Free the global opt[] (option list) when exiting.

These really are not of themselves important, but reduce the noise reported by
valgrind when looking for real issues.

(seeing a pattern here) :-)

Use weight == 0 instead of int == double when deciding to copy pixel
When interpolating width, do it by column, not row

I expect BitBlt(..., BLACKNESS) to be the better optimized of the
two, and there's no reason to actually reat the source bitmap.

Cleans up artifacts in interpolated values nicely.

I don't think it would've even been possible to match these expressions without
the tilde.

After a negative-match of a Guru expression, we were searching for the next
'(' anywhere in the subsequent chars of the file rather than just the next
line that started with an open-parenthesis character.

This bug was the secondary cause of the buffer-overflow crash addresseed in
commit 62e2280e which was triggered by my recent updates to the guru.dat.

Create a proper aspect fix function
Centre bitmap in window with correct aspect ratio
Fix cursor at resize points
Explicitly set high DPI awareness
Don't snap window larger than allowed

Since this is the newest/simplest display driver, I'm going to use
this to decide how to do scaling in general so I can abstract that
into shared code to ensure all drivers have the same scaling
behaviour.

I haven't tested mouse on windows with black bars yet though, so I
assume that's broken...

The 'list' member was allocated by iniParseSections() but never freed.

Caught be Nelgin's recent valgrind report, e.g.:

==2725235== 56 bytes in 1 blocks are indirectly lost in loss record 82 of 191
==2725235==    at 0x484DCD3: realloc (in /usr/libexec/valgrind/vgpreload_memcheck-amd64-linux.so)
==2725235==    by 0x4FA351C: strListAnnex (in /home/bbs/sbbs/repo/src/sbbs3/gcc.linux.x64.lib.release/libsbbs.so)
==2725235==    by 0x4F9ADDE: iniParseSections (in /home/bbs/sbbs/repo/src/sbbs3/gcc.linux.x64.lib.release/libsbbs.so)
==2725235==    by 0x49F1F0A: read_chat_cfg (in /home/bbs/sbbs/repo/src/sbbs3/gcc.linux.x64.lib.release/libsbbs.so)
==2725235==    by 0x49E8BF1: load_cfg (in /home/bbs/sbbs/repo/src/sbbs3/gcc.linux.x64.lib.release/libsbbs.so)
==2725235==    by 0x5C14C75: ftp_server (in /home/bbs/sbbs/repo/src/sbbs3/gcc.linux.x64.lib.release/libftpsrvr.so)
==2725235==    by 0x5D5AB42: start_thread (pthread_create.c:442)
==2725235==    by 0x5DEBBB3: clone (clone.S:100)

Will be needed once scaling works properly.

Load icon from resource file
implement seticon()

While the features are all there, some are buggy.  Scaling is wrong
for example, and keyboard input may do weird things if shift or
control are held when you take focus away from the window, then
release them when the focus is elsewhere, then re-focus the window.
This should only impact extended keys (ie: ones that don't send a
character to the remote).

Instead, only delete it if the size has changed since the last
frame.  This should make things snappier.