replace memcpy with AbstractMemory#put_bytes() in FFI implementation

[flavorjones]

From Wayne Meissner:

I noticed in nokogiri, in Document.read_io that you use LibXML.memcpy to transfer data from a ruby string to a native memory buffer.

It'll be much faster to use AbstractMemory#put_bytes(), as the memcpy mapping will triple copy the string data on JRuby.

basically, what happens behind the scene is:

1. The string data is copied to a temporary native memory buffer
2. The temp memory is passed as the parameter
3. memcpy copies the data from the temp memory to the destination memory you specified
4. The temporary native memory is copied back to the ruby string.

compared to put_bytes:

1. The string data is copied once from the ruby byte array backing the ruby string into the destination memory address.
2. There is no step 2.

For large files, this might make a difference.

[flavorjones]

FFI: IO reader callbacks now use AbstractMemory#put_bytes instead of memcpy, per Wayne's advice. along the way, refactored callbacks. closed by 0b87111.