Properly support \r and \n

which may be load()ed.

TODO: Make the calling of the function optional in case it's already been
load()ed

value for host_name comparison - under some strange circumstances, the
startup->host_name could be blank.

JS_ValueToInt32 fails - could this've been the cause of Deuce's mysterious
file.read() problems?

<DigitalMan> bad
<DigitalMan> and referencing file.length could very well reset the ferror()
value, so sample and save the file.error value immediately after error

forcing the log level to at least LOG_INFO when an item is read from the
.ini file.

This *should* result in the highest possible resolution.

to be able to accurately cast a long long int to a long double.  Replace:
(long double)x/y
WITH
(x/y)+((long double)(x%y)/y)

Can everyone look for a problem with assuming those two are interchangeable
with the second being more accurate?

Hopefully, this will fix the massively negative times.

are apparently the same thing, not so for nost *nix systems.  Unsure of BCC.

At the very least, MSVC doesn't support a long double type with *printf()
whereas *nix does.  This does not *seem* to be causing a problem, but it may.

method would insert keys at end of the input ring buffer. This was fine if
there weren't already chars in the buffer. But if there were already chars in
the input buffer, ansi_getxy() would then move keys to the end of the input
buffer until it found the <ESC>[y;xR response it was looking for. This garbled
the user's input in split-screen chat as reported by JasHud and others.
If there was a RingBufInsert() routine, we could use that instead, but this
separate "ungetkey" buf will do for now.

if Synchronet NT services were installed, but disabled.

<DigitlMan> the colons in the "Program Settings" menu are kinda ugly :-)

Make scrollback buffer shared.