Native coin tied-note release: sequencer verified, gameplay/listening pending

The local original coin effect3:17 contains finite note waits, tie commands and
an explicit END. It is not an embedded infinite sequence loop. The host player
created an indefinite tied voice and left it held when END marked the track
finished. A synthetic looping-instrument/tied-note regression reproduced this
before the fix. END now releases tied voices belonging to that track through
the existing instrument release envelope. Non-tied sample tails are preserved.
The generated Sound.cpp removes the old arbitrary60-game-tick coin timeout.
Matched source/header files are untouched; changes are under port only.

Build, synthetic regression, independent-track isolation, ordinary sample-tail
preservation and existing SDAT/mute tests pass. Local uncapped3:17 remains active
after10seconds before the fix and completes at1.067seconds afterward. Initial
0.73977seconds of PCM are identical; final seconds now silent. Local effects
3:18,3:331,3:332,3:334 produce byte-identical10second WAVs before/after. Their
completion times remain1.067,2.475,3.776,0.683seconds. No cannon timbre fix is
claimed. Full audit83music(10s each) and1448effects(2s each) passes without decoder
or execution errors; archive3's bank32 here is an explicit audit override.

Two muted original-menu gameplay checks exit normally with zero mixererrors,
but do not collect a coin: Battlefield route passes above the placed ground
coin row while climbing the bridge; castle lobby route reaches the wall below
the raised coin ledge. Those are incomplete pickup checks, not coin acceptance.
Stop guessing approaches; derive a route up the main staircase before traversing
the ledge, or use an already accepted input-only reward pickup fixture. Hearing
and complete DS audio fidelity remain unverified. Main F installation is not
promoted. Any separate diagnostic preview has private saves/settings.
