vmware: sync Fibre Channel HBAs of hosts as interfaces

[semx]

Fibre Channel HBAs of ESXi hosts as NetBox interfaces, for #343, behind sync_host_fc_adapters (default off).

For every vim.host.FibreChannelHba of a host (block and PCIe adapters are ignored) an interface is created on the device: the adapter's device name (vmhba3), the FC interface type derived from the adapter model's capability (QLE2700/QLE2800 32/64G … → 32gfc-sfp28; the port's current link speed is only used when the model names none, so a port that drops link keeps its type), the port WWN converted from the decimal the API returns to the 21:00:F4:C7:AA:9E:28:E6 form NetBox stores, model and driver as description, enabled from the adapter status.

The matching code needed one change: an FC HBA has no MAC, and interfaces that match nothing used to be paired 1:1 by name with whatever current interface was left. On a host that also had a BMC interface from the check_redfish source that pairing renamed the BMC interface to vmhba4 and moved its IP around. Now an interface with a WWN matches the current interface with the same WWN when no interface of its name exists (case-insensitive), and Fibre Channel ports, meaning an FC interface type or a WWN without a MAC address, take no part in the leftover pairing on either side. Interfaces with a MAC address are matched exactly as before, including converged adapter ports that check_redfish reports with both a MAC and a WWN; a test pins that. An FC adapter that reports no port WWN is skipped, since the WWN is what tells it apart from a NIC. A unit test reproduces the hijack and fails without the change.

Live NetBox 4.7.2 with the vc001 inventory (20 FC HBAs on 3 hosts) and check_redfish as a second source: the option on creates 20 FC interfaces with uppercase WWNs and types 32gfc-sfp28 (20 POST, no PATCH on existing interfaces or IPs), the BMC interface keeps its name and IP, and the second run writes nothing. (A first version sent lowercase WWNs, which NetBox normalises, and PATCHed every FC interface on every run; that is why the case matters.)

Tests: tests/test_vmware_host_fc_adapters.py (32): type derivation for every speed and for the model string, WWN conversion and padding, option off creates nothing, vc001 interfaces (exactly the 8 FC adapters and none of the SATA/NVMe controllers, types, WWNs, description, enabled state), the hijack, WWN matching when no interface of that name exists and across case, two HBAs sharing a WWN not merged, a current FC interface never handed to a new NIC, each of the two FC signals on its own, a MAC+WWN port still matched by MAC, host_interface_exclude_filter applied to the adapter names. On development without this change 28 of the 31 test runs fail; the three that pass pin today's behaviour (the MAC+WWN port, the exclude filter). Full suite: 311 passed, 1 skipped (development: 280). With the option unset the handler produces the same inventory as development on vc001: 324 objects, byte-identical canonical dump.