BBufferedDataIO: Fix Write().

I looked at this code as a reference when writing similar code in another
project, and realized then that this was broken to the point where I
don't think anyone could have actually used it:

1. Writes larger than the buffer should only return the amount written
by this Write() call, not the amount flushed beforehand.

2. fDirty was only set if there were bytes remaining after the first
write to the buffer. So, if you wrote small amounts of data, they
would be silently discarded most of the time!

3. When filling the buffer, we might as well flush simultaneously.
This allows the logic to be consolidated into a loop and remove the
duplicate memcpy. In case of failure, just return the bytes-written
as the write could be retried later.

4. Flush() should always return an error or 0, not bytes written.

While at it, add some basic tests for this class.

Change-Id: I2de01d0b31e3fe22863cef21dd7b0b62ed47121b

BBufferedDataIO: the underlying stream might return an error.

* In this case, fSize would have been set incorrectly, ignoring the error
completely, but open the possibility to read from uninitialized or even
unmapped memory.

BufferedDataIO::Flush(), use ssize_t rather than size_t

Write() was returing an ssize_t and Flush() was expecting a ssize_t,
but a size_t was instead used. CID 11052.

Added a buffered DataIO subclass.

* Only the read path is tested so far.

BBufferedDataIO: the underlying stream might return an error.

* In this case, fSize would have been set incorrectly, ignoring the error
completely, but open the possibility to read from uninitialized or even
unmapped memory.

BufferedDataIO::Flush(), use ssize_t rather than size_t

Write() was returing an ssize_t and Flush() was expecting a ssize_t,
but a size_t was instead used. CID 11052.

Added a buffered DataIO subclass.

* Only the read path is tested so far.