IOWaitLibrary ready? returns nil when ready

Test case passes in 1.0.3 and 1.1RC1 binary releases. However, in both of those releases, calling #ready? on a newly initialized socket with no available data on the stream causes it to hang.

After recent IO reorg, the IO classes are NIO channel-oriented, which brings one serious problem: no available().
Streams, created from the channels, provide trivial available() which always returns 0.

The ready? method actually just invokes available on such stream, and since it always gets 0, it always reports that no data available.

The Sun implementation of Channels.newInputStream does indeed seem to always return 0 bytes for available().

I put together a quick fix (io_wait_ready.patch) that creates a temporary Selector, makes the channel non-blocking, registers the channel, calls selectNow() to determine if there are bytes available, and returns to the channel's prior blocking mode. Someone with more knowledge of NIO should check this to make sure there are no ill side-effects. You can apply the patch to trunk/src.

This didn't turn out to fix my original problem with the STOMP gem, which I mistakenly thought was due to ready? always returning nil, so I'll do some more testing there. It turned out that ready? was not being called in the Stomp::Client class, since Stomp::Connection#poll is provided but not used...

I committed this and posted it for review on dev list, since it seems a little heavier than I'd like. If it's possible to get a test that doesn't depend on making an internet connection, that would be great.