Compensate for the cache delay, and actually populate the cache.
Without these, the playback would start with garbage (which would be
(mostly?) masqueraded by the attack phase).

Unlike for the PCM voices, this doesn't try to compensate for the
interpolator read-ahead, because it's pointless to be super-exact here.

Note that this code is probably still broken for particularly short
samples, because we ignore the loop-related parts of CCR. But I'm not
going to reverse-engineer that now ...
Signed-off-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Link: https://lore.kernel.org/r/20230518140339.3722308-1-oswald.buddenhagen@gmx.deSigned-off-by: Takashi Iwai <tiwai@suse.de>