Report playback from the null and stdout sinks so the player releases chunks

[chrisuthe]

The null and stdout sinks never reported playback, so the player's sync task stayed in initial-sync priming: the codec header stayed held, every chunk queued behind it in the inbound ring, the player's quota overflowed, the library closed the connection, and the server reopened it in a loop. Both sinks now report every write as played against a virtual clock, so each chunk is released as it arrives.

Fixes #82

What changes

• NullAudioSink::write() reports the frames it consumed, in the discard, muted and stdout branches alike, with a finish timestamp on a playhead that is never behind real time: max(now, playhead) + frames / rate. The clock runs ahead of real time by whatever the server sends ahead, which is what lets the sync task release each chunk on arrival; it is the pattern the library's own VirtualSinkListener uses. A short stdout write reports what it consumed, so the sync task's frame accounting does not drift.
• The audio_sink.h note that said reporting was optional for instant sinks is corrected: a sink that never reports stalls the player before its first chunk.
• tests/null_sink_test.cpp: every write is reported with its frame count on a clock that keeps advancing; nothing is reported before configure(), for a zero-length write, or after stop().

Verification

Against Music Assistant 2.11.0.dev with a 44.1 kHz / 24-bit FLAC track, the same player identity on the same server, -o null -d debug.

main before this change, three stream starts in a row:

I player: Stream started
I sendspin.sync_task: Processed new codec header: flac, 44100 Hz, 2 ch, 24-bit
W sendspin.inbound: Player over its buffer; dropping items until it drains
W sendspin.connection: No inbound ring space for a 17911-byte message within 100 ms: ring pinned behind held items (986344 bytes held); closing the connection
D sendspin.conn_mgr: Admitted connection closed by the client
W sendspin.inbound: Dropped 13 player items
I player: Stream ended

With this change, the player restarted on the same state directory while the server kept playing:

I sendspin.conn_mgr: Connection admitted: server_id=OraobU4leb48EaZTGPbLhNug8XjEZsKXv8qiczhEGTU
I player: Stream started
I sendspin.sync_task: Processed new codec header: flac, 44100 Hz, 2 ch, 24-bit

One stream start, no warning of any kind, 0.2% CPU over the following minute.

ctest 521/521 apart from ConfigMerge.AConfiguredControlSocketIsAbsolutizedUnderDaemonize, which fails only from a 105-byte worktree socket path; scripts/format.sh --check, scripts/smoke_test.sh and a -DSENDSPIN_CLI_WERROR=ON build are clean. Device-backed sinks already report and are unaffected.