shared/table: do not pass a NULL string to strdup()

[mittal-ishaan]

nvme list segfaults when a controller reports a blank firmware revision.

libnvme's __nvme_get_attr() strips trailing blanks and returns NULL when nothing is left, so libnvme_ns_get_firmware() returns NULL for such a controller. list_item() hands that to shr_table_set_value_str(), which calls strdup(str) unguarded and dies in strlen().

Seen on Ubuntu 26.04 (nvme-cli 2.16, libnvme 1.16.1), GCP node whose persistent disk (model nvme_card-pd) has an all-spaces firmware_rev:

~# nvme list
Segmentation fault         (core dumped) nvme list

~# nvme list -o json | jq '.Devices[] | select(.Firmware == null) | .DevicePath'
"/dev/nvme32n1"

A NULL string is now stored as an empty string, so the cell stays blank. Returning -EINVAL instead would make the callers drop the whole device row. The same setter also feeds serial, model and controller address, which can be
NULL for the same reason.

[igaw]

I am not against making the table code more robust, that's fine.

Though we could just do the right thing with the defaults, that means instead of

libnvme_ctrl_get_firmware(n->c, &val, NULL);

do a

libnvme_ctrl_get_firmware(n->c, &val, "");