support (seekable) ByteArrayInputStream#to_io.rewind

[kares]

except FileInputStream it seems quite useful to be able to rewind/seek a byte[] backed stream

this is not possible using Channels.newChannel as the returned channel (except for files) is not seekable
but using the 9K refactored RubyIO its fairly easy to play a trick by returning a custom channel instance ...

1.7.x remains behaving weirdly on rewind ... more notes at #2381

[headius]

@kares I appreciate the attribution but the code at 6142ea3#diff-b60adbaec9c15033c4eb0231a09d211aR46 worries me. We can't incorporate code from GPL codebases like this, even if they're been massaged into place a little.

I'd suggest you look at the Android implementation of this logic, which is licensed Apache-2.0. We can incorporate them into a separate file with an Apache license header and it should be compatible with all three of our licenses.

Here's the Android 4.3 impl of Channels at equivalent read method: https://android.googlesource.com/platform/libcore/+/android-4.3_r1/luni/src/main/java/java/nio/channels/Channels.java#297

[kares]

@headius thanks, the problematic part is now re-implemented as suggested.
tests fail due missing jnr-posix SNAPSHOT (currently, there's no green travis-ci build to rebase this off).

[headius]

@kares I'll pull it down today, review and make sure it's building, and push it out.

[headius]

Testing seems fine, I'll push shortly.

Last suggestions, if you get a chance (just throw them on master):

• Try to grab and setAccessible the fields you access during static initialization; if we can't do it, we need to fall back to the unseekable wrapper.
• Saving the fields will also be better perf rather than looking up and hitting setAccessible every time they're accessed.