pcm_meter: let the s16 scope degrade instead of killing the caller

[Gjuju]

The bug

s16_enable() returns -EINVAL for any format it cannot convert to S16 — the DSD
formats, the 3-byte packed ones (S24_3LE/S24_3BE/S20_3LE), float.
snd_pcm_scope_enable() records that as "not enabled" and carries on:

err = scope->ops->enable(scope);
scope->enabled = (err >= 0);

But every other scope stays enabled, and scopes reach the s16 buffer through
snd_pcm_scope_s16_get_channel_buffer(), which asserts on exactly the pointer
s16_enable() never allocated:

pcm_meter.c:1224: snd_pcm_scope_s16_get_channel_buffer: Assertion 's16->buf_areas' failed.

The application dies. This is not hypothetical and not third-party only: alsa-lib's own
src/pcm/scopes/level.c calls the accessor the same way, so the library aborts its own
caller. Downstream, libpeppyalsa (the VU-meter scope used by moOde audio player) makes
a music player abort mid-playback the moment a DSD track starts.

Note the failure is not DSD-specific — any format outside the 17 the switch lists
triggers it, including S24_3LE, which plenty of USB DACs expose natively. A chain that
happens to have a plug in front of the meter is merely lucky.

The fix

Allocate the buffer anyway and leave it zero, so a scope reads silence on a format
the s16 conversion cannot see. Callers need no change, and none can be made to abort —
which matters, since returning NULL instead would only move an unfixed caller from
SIGABRT to SIGSEGV. The condition is now reported once through SNDERR() instead of
being swallowed.

Testing

x86-64, Debian 13, USB DAC with a native DSD path, moOde audio player with its VU meter
scope in the chain. Playing a DSD64 .dsf with dop=no:

• before: MPD aborts on the assertion, the UI reports "Socket open failed"
• after: the DAC receives DSD_U32_BE @ 88200 bit-perfect, MPD stays up
  (NRestarts: 0), needles read zero, and the log carries
  s16 scope: no S16 conversion for format DSD_U32_BE, scopes will read silence

Left alone

The neighbouring S16 + MMAP_NONINTERLEAVED branch assigns s16->buf and then also
returns -EINVAL, reaching the same dead end. It looks unfinished rather than
deliberate, so this patch does not touch it — happy to follow up if you can confirm the
intent.