This allows running syncterm from cmd.exe using just "syncterm"
to run in console mode, but using the .exe to run in Windows mode.

It should now be possible to run SyncTERM in ANSI or console mode
inside of SyncTERM in GDI mode.

Fixes bug where you could only open a Shell session on Windows once
per invocation of SyncTERM.

Requires Windows 10 or higher.  This changes the default SyncTERM
builds to require Windows 10.  To build for other versions, you
need to pass WIN32, WINVER, WIN32_IE, and NTDDI to build script.

These need to be whatever they need to be for old stuff to work.

- Make lzh_update() return a bool indicating success
- Eliminated calculating array size, use macro
- Fail if the current entry is not in the frequency array

warning: adding 'int' to a string does not append to the string
[-Wstring-plus-int]

Another fix for bug 174

cryptPopData() and cryptFlushData() will return CRYPT_ERROR_READ
if the socket is unexpectedly closed, not CRYPT_ERROR_COMPLETE.

Handle that in the wrappers as well.

If anything that should send/receive data returns CRYPT_ERROR_COMPLETE
tell the I/O threads to temrinate so conn_connected() will return false

Another attempt at bug 174.

Historically, casting to void would suppress warnings about things
being unused.  However, an attribute was created for functions like
realloc() where not using the result is always a bug, so when the
attribute is used, you must actually use the return value and can't
cast to void to avoid a warning.

Given this attribute, the glibc authors decided to apply it to all
the functions that usually should have their return value checked
(chdir(), write(), etc) when _FORTIFY_SOURCE is defined because if
you care about correctness enough to define _FORTIFY_SOURCE, you'll
always do something useful with these return values.

Extending the hilarity, some distros define _FORTIFY_SOURCE by
default.

The result is functions that occasionally can safely be called
regardless of the return value will throw a warning if you don't
"do something" with that return value.

Of course, you can just turn off the warnings completely with a
compiler flag, but then you lose the excellent warning about things
like realloc() that always make sense and were the original reason
for the option.

Pretty much all new compilers these days will define all the GCC
macros so it's impossible to detect GCC from predefined macros
without adding a list of all the compilers that pretend to be GCC
if defined(__GNUC__) && !defined(__clang__) && !defined(____INTEL_COMPILER)

Overhaul LZH code

See merge request !489

Fixes memory leak found by scan-build.

Found by scan-build

Found by scan-build

Found by scan-build

This would be an API violation, but we shouldn't crash about it.

Prevents randomly sized windows in X11 mode when the max window
size can't be read from window manager.
Found by scan-build

Prevents X11 mode from doing very weird things when we are unable
to get the max window size from the window manager.
Found by scan-build

Previously, this could (but likely wouldn't) result in accesses
outside of allocated memory, doing Very Bad Things.
Found by scan-build

In the CSI 38 : 2 : Z? : R : G : B m variant, if there was no
sub params (ie: CSI 38 : m), a read out of bounds would be
triggered.
Found by scan-build

Fixes potential read of uninitialized data.
Found by scan-build

If strbuf is over 5MB, or the realloc() fails, clear strbuf and
break before writing to strbuf.
Found by scan-build

That's just silly.
Found by scan-build

Shouldn't happen, but if it does, return an error, don't dereference
Found by scan-build.

Caught by scan-build.
Not actually possible with any of the current backends, but the API
would allow it to happen.

fix for commit 6e69eec1

This especially important/noticeable on a 4K monitor.

Previously, it wrote to the last cell on the screen, so would
detect column 1 in terminals that wrapped, now moves back one
cell before printing the _.

Also, since that's fixed, grab the number of columns as well.

This can't actually happen.