cxl/core: Replace devm_cxl_add_decoder() with non-devm version

cxl/core: Replace devm_cxl_add_decoder() with non-devm version

The split of devm_cxl_add_decoder() to factor out cxl_decoder_alloc()
introduced a couple bugs. An initialization order bug, and a layering
violation around assumptions about who is responsible for put_device()
when device_add() for the decoder fails. Fix this by making the caller
responsible for registering a devm callback to trigger
device_unregister() for the decoder device.

Fixes: b7ca54b62551 ("cxl/core: Split decoder setup into alloc + add")
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Nathan Chancellor <nathan@kernel.org>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: https://lore.kernel.org/r/163002073312.1700305.17017280228713298614.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>