If the decoder prepare_run callback returns an error, the hardware
is never started and v4l2_m2m_job_finish() is called immediately,
currently in every error path.

Remove the duplicated job_finish calls from the error paths and reuse
the v4l2_m2m_job_finish() at the end of coda_pic_run_work instead.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>