remoteproc: qcom: wcnss: Fix race with iris probe

The remoteproc driver is split between the responsibilities of getting
the SoC-internal ARM core up and running and the external RF (aka
"Iris") part configured.

In order to satisfy the regulator framework's need of a struct device *
to look up supplies this was implemented as two different drivers, using
of_platform_populate() in the remoteproc part to probe the iris part.

Unfortunately it's possible that the iris part probe defers on yet not
available regulators and an attempt to start the remoteproc will have to
be rejected, until this has been resolved. But there's no useful
mechanism of knowing when this would be.

Instead replace the of_platform_populate() and the iris probe with a
function that rolls its own struct device, with the relevant of_node
associated that is enough to acquire regulators and clocks specified in
the DT node and that may propagate the EPROBE_DEFER back to the wcnss
device's probe.

Reported-by: Anibal Limon <anibal.limon@linaro.org>
Reported-by: Loic Poulain <loic.poulain@linaro.org>
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Tested-by: Anibal Limon <anibal.limon@linaro.org>