fix: port_init_port_id silently uses wrong port_id when PCI lookup fails

[digger-yu]

Root cause:
port_init_port_id walks the per-port pci_list and calls
rte_eth_dev_get_port_by_name for each entry. When the
lookup fails, the function prints a warning and falls back
to assigning the loop index i as the port_id, then continues
the loop. The loop index is not a DPDK port_id; DPDK port_id
is assigned by DPDK at startup based on the PCI allow list
and any --vdev entries, and is not guaranteed to match the
user-provided pci_list ordering.

Impact:
When a pci_list entry does not match a DPDK-managed device,
the function stores a fabricated port_id into
port->port_id_list[i] and returns 0. Subsequent calls that
use port_id_list[i], including rte_eth_dev_socket_id and the
rx/tx init calls, then operate on whatever DPDK device
happens to occupy that numeric id, which can be a different
port or no port at all. The warning text makes the issue
look benign, so a misconfigured dperf can keep running and
emit corrupted traffic or hit unrelated device errors far
from the root cause.

Style consistency:
The fix removes the fabricated fallback by replacing
port_id = (uint16_t)i with return -1, matching the fail-fast
pattern used throughout the rest of port_init and across
config_parse_* in config.c. No other call sites or files
are affected.