* convpath logics implemented properly now.
* Most other methods still to be converted.

* From my impl (modified) at rubinius/rubinius#2912
* Fails due to ArrayStoreException in jcodings

* Moved ECONV flags to use those from jcodings
* Many port fixes and tweaks.

Nearly there!

…fying the class is not a Module.

Revert "Update tests from MRI 2.1 branch."

This reverts commit 9c9187c.

* More exact port of rb_reg_regsub
* Reverted fix to ASCGET logic that breaks a test; unknown reason
* Expanded strBufCat and related methods to take byte[].

This avoids performing the potentially expensive hash computation in
the event that the Hash is empty, as there cannot possibly be a match.

In the event that the `#hash` method has side effects, any optimisation
with respect to empty hashes will affect the results of the operations.

For MRI, the empty-hash optimisation seems to be applied everywhere in
2.0.0 and beyond. In 1.9.3 the #hash method is not called for #[] and
#key?, but strangely enough the performance is not improved for empty
hashes. In 1.8.7, the #hash method is always called even if empty.

This does not match the semantics of MRI exactly, as this avoids calling
the #hash method from #values_at in 1.9.3, but that would be slightly
more involved to implement.

Conflicts:
	core/pom.xml
	core/src/main/java/org/jruby/RubyFile.java
	core/src/main/java/org/jruby/util/TypeConverter.java

This logic forced me to implement all the MRI functions for doing
buffered, possibly-transcoded reads from a stream. I also had to
implement some of the write logic.

As other IO methods all out, we should start to move away from our
flawed ChannelStream abstraction and toward the MRI way. Later,
we can look at doing a better abstraction of the MRI way.

What's wrong with ChannelStream, you ask? Nothing, provided all
you want to do is POSIX-like FILE* IO. MRI stopped using FILE*'s
style buffering at some point -- probably when adding character
transcoding -- and as a result we started mixing encoding logic
into ChannelStream. This caused the code to become rather confused
and no longer useful as a bare FILE* mechanism...while also
preventing us from implementing IO+transcoding as efficiently as
MRI (which does unbuffered reads into its own buffers and
transcodes in place).