When passed null or undefined, it needs to return null or undefined.

Document that behavior.

It's CryptCert.FORMAT.XXX

Add a nextPacket() method that will get the next packet
Add a Connection.parseBytes() method to consume the rx_buf
Simplify Packet.newBytes() to only do socket recv()
Clear RX buffer when closing socket

- Increment bytes required for every extra VBI byte we learn about
- If sock is null, don't look at sock.connection
- Set sock to null after close
- Don't read from sock unless we need some bytes
- If we already have some bytes, parse them

There's a couple hacks in here that should be refactored though.

This appears to be why it was hanging. :D

This is not nearly ready for full-time yet, but it seems to mostly
limp along fine... there's some issues with the callbacks and such,
and the TLS support isn't currently usable by Synchronet.

Worth committing though.

Internal codes are always treated as keys (in a database sense) and should be
unique, but they weren't always enforced to be unique in SCFG. There's still
a few config items with internal codes that aren't forced unique (e.g. chat
channels).

File Library short names and Mesage Group short names are also key values and
must be unique and that's now enforced as well.

Remove the default name for new File Libraries and Message Groups (which are
likely duplicate anyway).

Used in combination with K_EDIT, if the string isn't actually changed, return
-1 (same as abort/ESC).

This allows the elimination of some save/compare/copy code in SCFG where we
only care to even look at the string if it was changed from its existing
value.

Broken when I removed the ALT-X shortcuts for ANSI/Curses modes.

Should fix issue #188

A couple were valid issues, but the majority were impossible.

20ms, so you wouldn't notice, but it may be the cause of problem
reports in the future, so best to have it logged somewhere.

Should "fix" the timing difference in login info between modes.
This removes a 10ms SLEEP() between the different pieces of data,
and would likely have resolved ticket 185 if it wasn't already
closed.

These prefer the protocol after the - and only include one listing
per BBS entry.

Default is SSH, then Telnet, then RLogin
The most preferred one is the one that doesn't get the protocol
as part of the BBS name.

Also, add an "onlyone" parameter (defaults to false) that requests
each BBS only gets one entry.

This makes watching the valgrind tests more entertaining.
Likely makes it easier to tell what test crashed too, but let's
face it, I'm doing this to be entertained.

This doesn't fix anything since calling exit() from a DLL is
nonsense, but at least it's less wrong than just waiting forever.

Also, when main() returns, tell the frontent to shut down.

It's most likely that this ends up causing the frontend to crash,
but if the stars align, it may exit instead.

Some BBSs (ms.bsrealm.net) have a mailer running on the initial
socket, and telnet negotiations with that disconnect.  Other
systems could potentially perform the negotiation then fail to
pass the telnet state to the next program.

With the new "Defer Negotiate" telnet option, SyncTERM won't send
any telnet negotiations until it receives a telnet command from
the remote first.

Win32CIO already does that.