decoder_load_replay_gain() will create the InputStream anyway, so
we're not saving any overhead by opening the InputStream on demand
only.

Prepare for APE tag support on userspace NFS/SMB/CIFS mounts.

Prepare for ID3 support on userspace NFS/SMB/CIFS mounts.

Use a 128 byte buffer instead of reading 10 bytes first and then 118.

The check was added in commit 98fd9b7d, but there was never a
definition.

Obsolete class CloseSongEnumerator, which was a kludge.

Copy the query buffer to the allocated buffer, and read only the
remaining data.

Omit the file size check if we don't know the size.