wireguard_auto_client hardcodes IPv4 family/mask, and the ipv6 schema branch rejects template variables

Description

OpenWrt.wireguard_auto_client() hardcodes the client address as IPv4:

# netjsonconfig/backends/openwrt/openwrt.py
if data["client"]["ip_address"]:
    config["interfaces"][0]["addresses"] = [
        {
            "proto": "static",
            "family": "ipv4",
            "address": data["client"]["ip_address"],
            "mask": 32,
        },
    ]

When the WireGuard VPN uses an IPv6 subnet, the generated client address is
still declared family: ipv4 with mask: 32, so the rendered configuration is
wrong for every consumer of this helper. server_ip_network is already passed
in through **kwargs, so the family can be derived rather than assumed.

This is the root cause of openwisp/openwisp-controller#721. A workaround was
proposed controller-side in openwisp/openwisp-controller#1255, and during review
@nemesifier noted the fix really belongs here so every consumer benefits.

Second, related problem

The controller-side workaround cannot actually work, because of an asymmetry in
the WireGuard interface schema. The addresses items use a oneOf with an ipv4
and an ipv6 branch. The ipv6 branch carries "format": "ipv6"; the ipv4 branch
deliberately has no format:

# netjsonconfig/backends/openwrt/schema.py, wireguard_interface
{"title": "ipv4", ... "address": {"type": "string", "minLength": 7, ...}},
{"title": "ipv6", ... "address": {"type": "string", "minLength": 3,
                                  "format": "ipv6", ...}},

BaseBackend.validate() runs Draft4Validator(..., format_checker=format_checker),
so the format is enforced. The WireGuard client address is normally a template
variable that is only resolved at render time, for example
{{ip_address_8841b49ed8344a0c8fb35f8286eccb41}} in OpenWISP. That value passes
the ipv4 branch and fails the ipv6 one.

Steps to reproduce

from netjsonconfig import OpenWrt

base = dict(name="wg0", type="wireguard", private_key="{{pvt_key_x}}", port=51820,
            mtu=1420, nohostroute=False, fwmark="", ip6prefix=[], network="")

def check(addr, label):
    try:
        OpenWrt({"interfaces": [dict(base, addresses=[addr])]}).validate()
        print("PASS ", label)
    except Exception:
        print("FAIL ", label)

check({"proto": "static", "family": "ipv4", "address": "{{ip_address_x}}", "mask": 32}, "ipv4 + template var")
check({"proto": "static", "family": "ipv6", "address": "{{ip_address_x}}", "mask": 128}, "ipv6 + template var")
check({"proto": "static", "family": "ipv6", "address": "fd00::1", "mask": 128}, "ipv6 + literal address")

PASS  ipv4 + template var
FAIL  ipv6 + template var
PASS  ipv6 + literal address

Isolating the failing keyword:

branch 0 (ipv4): enum='ipv6' is not one of ['ipv4']
branch 1 (ipv6): format='{{ip_address_x}}' is not a 'ipv6'

Proposed fix

1. Derive family and mask in wireguard_auto_client() from
   server_ip_network, defaulting to IPv4 when it is missing or unparsable so
   current behavior is preserved.
2. Drop "format": "ipv6" from the WireGuard ipv6 address branch, matching the
   ipv4 branch, since these addresses are routinely template variables.

I have both changes working with tests and can open a PR.

Start in netjsonconfig/backends/openwrt/openwrt.py at wireguard_auto_client(), then inspect the wireguard_interface address branches in netjsonconfig/backends/openwrt/schema.py. Run the supplied validation examples and compare IPv4, IPv6, literal, and template-variable behavior. Done means IPv6 client settings derive the correct family and mask, and IPv6 template variables validate like IPv4 ones.