Both selecting scrollback from the online menu, and viewing the last
scrollback from the main menu were impacted.

Currently, this is a slower way to look-up a text.dat string, but requires no
load/require of text.js and allows for a shorter syntax to get a text.dat
string, i.e. bbs.text("Quit") instead of bbs.text(bbs.text.Quit);

Although the bbs.text() ID lookup is cached, it's only marginally faster than
system.text() ID lookup, which is not cached and about 1/3 the speed of the
other methods of index to string lookup (in the MSVC-Windows build).

... with Servers->Terminal Server->Login Requirements

Updated help text: F5->Ctrl-C, no need to hand-update sbbs.ini when adding
or removing nodes (any more).

- Attempt to suppress go aheads in both directions.
- Attempt to set binary mode in both directions
- Request the remote echos

These express the assumptions that SyncTERM always makes, so it's
a good idea to tell the remote this.  For systems that don't like
this, there is the Raw "protocol".

When changing between 8 and 16 row fonts in RIP mode, the cursor
start and end was left for the old font size... 16 -> 8 would leave
the cursor one line below the current position, and 8 -> 16 would
leave the cursor in the middle of the cell.

This hacks the vstat deeper to fix up the cursor as well.

Reported by skipperdoodle1947.

This is a code clean-up, no change in functionality

It's possible to have multiple transfer protocols with the same mmemonic, but
with different transfer types supports (which would be weird, but possible).
This allows us to use protnum() to replace a lot of copy/pasted prot looping
and comparing logic throughout sbbs.

Also added a variant of sbbs_t::quit_key() that takes and returns a string
(for easy concatenation to strings of key chars).

If an ANSI sequence spanned multiple recv() calls, it would get
silently dropped if RIP was enabled.

Reported by skipperdoodle1947 (who does awesome stuff)

for use by JS bbs.xtrn_prot_menu() immediately, elsewhere soon.

... that the user has access to

So we don't have to reimplement this logic in user_settings.js (or
equivalent) any longer.

Also, fix manpage generation for CMake.

I've forgotten one of these for every RC except the first one for
v1.2. :(

Thanks nelgin.

The amazing Haiku community has redesigned the SyncTERM icons for me!

Unfortunately, I had to remove the default "Synch" icon from ciolib
so that SyncTERM could have its own separate icon. :(

I'm not sure if the msvc builds are using the ciolib.rc file or not,
and I'm not sure how they can add an icon... I'll net DigitalMan sort
that out (sorryish).

This was sometimes causing an integer scaled mode (ie: 4×) to be
reduced to very slightly less than that (ie: 3.996255×), making the
window one pixel smaller than it was supposed to be due to integer
truncation.

This could result in being unable to increase the window size using
the Alt+Right-Arrow shortcut and could make new windows one full
size smaller than they needed to be (such as C64 windows).

While we're here, fix the SDL and X11 outputs to trust what
bitmap_drv_init_mode() does.  It was fixed when GDI mode was being
written, but the other drivers weren't updated to take advantage of
that.

Should resolve issue 156 reported by DigitalMan

Whee.

Use localizable [Q]uit key

Removed trailing whitespace

Now it doesn't leave the console all messed up.

Helpful for a JS batch file transfer menu

Returns count of files removed now (not bool).

This allows GetConsoleScreenBufferInfoEx() to work for palette
changes, allowing Console to not suck.  However, we now prefer
ANSI mode over Win32 Console, so make that so it can actually
fail on Windows of stdout is a TTY and we can't set the
ENABLE_VIRTUAL_TERMINAL_PROCESSING flag.

With this, ANSI mode is used in Windows Terminal, and Win32 Console
is used in Legacy Console, and all should be good with the world.

CSI s and CSI u are ANSI.SYS extensions, and aren't needed here.
We're clearing the screen and moveing the cursor anyway.

Not only was it broken when window() is set, it's also broken when
the width is not the full screen width.

We've now disabled pretty much all the optimizations... but it works.

Specifically, make Win32 Terminal output not suck.
If stdout is a tty, set the palette appropriately, avoid using \n
since Windows can't seem to be stopped from expanding it to \r\n
nd brob the Console modes a bit harder.

The window and the screen buffer are two different things... the
screen buffer must always be the same size as or larger than the
window.  Further, there are times where it's not possible to resize
the windown programatically (new terminal for example).

There still appears to be a stupid bug in what I assume is the
Windows Terminal LCF flag implementation which causes writes to
the beginning of a line that occur after a write to the end of the
previous line to be in the wrong position.  If it's not the LCF flag
(which isn't turned on), it's in their wrapping thing which is extra
irritating since I explocitly turn that off.

Almost there, the main issue remaining is when running from a
command-line, it sometimes "unwraps" lines.

Likely still not right though.

Because it screws up the display when you resize the window.