When the client wants to seek, but the decoder has already finished
decoding the current song, the player restarts the decoder with an
initial seek at the new position.  When this initial seek fails, MPD
pretends nothing has happened and plays this song from the start.

With this new flag, a restarted decoder marks the initial seek as
"essential" and fails the decoder if that seek fails.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/895

Previously, a bogus value (whatever happened to be still in
`start_time`) was used.

Fixes formatted paths in the `recorder` output, closing #345.

This tiny amount of overhead allows omitting the MusicBuffer in
Clear().

This adds a bit of overhead, but also adds flexibility to the API,
because arbitrary triggers may be invoked from that virtual method
implementation, not just Cond::signal().

The motivation for this is to make the handlers more dynamic, for the
upcoming buffering class utilizing ProxyInputStream.

If an early exception gets caught (e.g. from
AllocatedPath::FromUTF8Throw()) before
DecoderControl::CommandFinishedLocked() is called, the decoder thread
would go in an endless loop, because DecoderCommand::START is still
set.

Closes #118

This eliminates some overhead, because the compiler doesn't need to
consider these functions throwing.

This repairs the damage to commit 74dbaade done by commit
b3f5b493

 https://bugs.musicpd.org/view.php?id=4595

Catch ENOTDIR exceptions and in that case, try all matching decoder
plugins which implement the "container_scan" method.

 https://bugs.musicpd.org/view.php?id=4561

Replaces decoder_data() and others.

Code simplification.

Be exception-safe.