ASoC: msm-lsm-client: free sound model buffer in case of ADSP SVA

Sound model buffer is allocated and mapped to DSP during sound model
registration. This memory is not unmapped and freed during deregistration
of sound model in case of ADSP based SVA. Fix this issue by calling
q6lsm_snd_model_buf_free() after sound model deregistration.

Change-Id: I8ffe3b4c26cdf90c716bf7f68435712dd8de6950
Signed-off-by: Vidyakumar Athota <vathota@codeaurora.org>