Kernel 4.4: use fwnode for irq domain

When creating an IRQ domain, the code used to store a device struct
pointer in the `of_node` field of the irq_domain struct returned by
`irq_domain_add_linear`. This was done by casting to a void pointer and
passing as the first argument of the function.

In the SPEC repository (and supposedly other depending repositories),
we would then get back the irq_domain struct by calling `irq_find_host`
with again a device struct pointer casted to a void pointer as the
argument. The function would compare the addresses of the 2 device
structs and return the right irq_domain.

This trick was most likely due to the fact that the IRQ domain API was
conceived around OpenFirmware before version 4.4, and that the project
should work for ACPI.
A workaround for kernel 4.7 was written, which involved using a `select`
function, and passing the address of the device struct as a parameter.
It was particularly ugly as it would require getting around the
`irq_find_host` to call immediately `irq_find_matching_fwspec` and pass
in the address of the wanted device struct as a parameter in a hacky way
to fit it inside two 32 bit integers.

Kernel version 4.4 introduced fwnodes, which would make easier working
with ACPI. Instead of repeating the hacky workaround (which would result
in a kernel error on later kernels when calling irq_domain_add_linear),
I allocated a fwnode_handle struct before creating a new IRQ domain.
Then I used the new irq_domain_create_* API to get an irq_domain using
this fwnode. I also took care of disallocating the fwnode_handle.

On the other end, we can just call `irq_find_matching_fwnode` and pass
dev.fwnode as a parameter.
Signed-off-by: Gwenhael GOAVEC <gwenhael.goavec@femto-st.fr>
Signed-off-by: Mathis MARION <mathis.marion@grenoble-inp.org>