It doesn't make sense to calculate the delay for capture streams in
the current implementation.  It's always zero, so we should skip the
computation in snd_usb_pcm_pointer() in the case of capture.
Signed-off-by: Takashi Iwai <tiwai@suse.de>