engine: improve universe composition efficiency

[mcallegari]

• Universe::tick() now releases the semaphore only if no pending token exists.
• Elapsed-time aware fading: improves fade timing accuracy under variable CPU load and scheduler jitter; fades progress according to real time instead of assuming perfect 20 ms cadence
• Reduced lock hold in Universe fader processing to avoid holding the universe fader list lock during potentially expensive per-fader channel writes, reducing contention and improving scalability with many faders/universes
• Lower allocation/copy pressure in output emission: avoids per-tick temporary copy for plugin output write path while preserving safe ownership semantics for queued signal consumers
• GenericFader channel lookup/update restructuring: reduces lock churn and repeated channel metadata setup overhead in high-channel scenes
• Locking correctness improvements in GenericFader: fixes data-race risk and ensures thread-safe mutation behavior
• Additional performance validation is done through engine/test/universeperf

[coveralls]

coverage: 34.05% (+0.08%) from 33.973%
when pulling 82d3585 on unirework
into 997835b on master.