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

Document that behavior.

Defeat the Windows getaddrinfo feature:
If the pNodeName parameter contains an empty string, all registered addresses on the local computer are returned.

at least for me, "fe80::5dfc:b29c:85d6:8de7%21"

at least, "vert.search.charter.net" in my case

Some scripts (e.g. addfiles.js) depend on the previous (undocumented) behavior
for some methods, e.g. lfexpand(undefined) would return undefined.

So for global functions that are expected to return a modified version of the
value passed (usually a string), return null or undefined when passed null or
undefined. Backward compatibility is more important than consistency in this
case. :-(

The behaviour is unreliable in Spidermonkey 1.8.5, and the behaviour
will change based on JS options.

This is really an error condition, so like the jsdocs say:
return undefined upon error

This how read(0) worked in sbbs, but not jsexec before that last commit,
so now just make readln and read work like that in both sbbs and jsexec.

Many global functions now throw an exception when null or undefined is passed
as a required parameter where null or undefined makes no sense.
e.g. "Argument #1 is an unexpected 'null' or 'undefined' value"
Example exception: is-a/has-a type functions will still return false when
passed null or undefined.

Changed global function return types:
- write() now always returns undefined
- printf() always returns a string, never undefined or garbage string
- prompt() returns null (not undefined) upon no input (as documented)

sbbs works more like jsexec:
- read(0) now returns an empty string instead of undefined
- readln(0) no longer waits for enter

jsexec works more like sbbs:
- add missing write_raw() global function, alias for write()

Include expected number of argment in "Insufficient Arguments" exception
message.

Update JSDOCS for some global methods:
- file_chmod returns boolean, not number
- flags_str requires a minimum of one argument

New tests to validate expected exceptions and return types based on usage.

The first argument after the list was still being replaced.
Extended test to try the hard stuff.

at least three times since we enabled the test suite.  The odds
of that happening *should* be one in a million, but NetBSD managed
to fail this time (don't remember previous times, and not sure
how to find them).

For now, I'll just change this so there's a one in a trillion chance.

Various bits needed to be hacked up a bit since it was written for
a newer JS standard, but it's at least 11 better than 42.