When reading from a decompressor that was constructed using `Decoder::with_buffer`, the decoder may not consume all of the input by the time it returns all of the decompressed output.  This means when you call `.finish()` on the decoder to get the underlying stream back, it is not pointing after the end of the compressed data.

This commit adds a test that demonstrates the issue.

After the decoder stream has yielded all of the uncompressed data, it is possible for the input stream to still not be fully consumed.  This means if we extract the inner stream at this point, it will not be pointing to the end of the compressed data.

From the [zstd documentation](https://facebook.github.io/zstd/zstd_manual.html#Chapter9) for `ZSTD_decompressStream`:

> But if `output.pos == output.size`, there might be some data left within internal buffers.
> In which case, call ZSTD_decompressStream() again to flush whatever remains in the buffer.

This is only necessary if the caller wants the stream back, so at that point we can force an additional call to `ZSTD_decompressStream` by reading to a zero-length buffer.