templates: fix addrman patch

[pinheadmz]

The current addrman patch results in an addrman that buckets ipv4 address by 32 bits instead of 16. The issue is that in NetGroupManager::GetGroup() we are dealing with IPv4 mapped to IPv6 addresses and so the 32 bits the function actually processes are the null prefix in the embedding scheme. This means all ipv4 addresses in warnet always end up in the same bucket :-P

The solution in this PR ensures that we process the last 32 bits of the embedded ipv6 structure which is where the actual ipv4 address really lives.

The issue is illustrated really well in the debugger while stopped in GetGroup():

(lldb) print address
(const CNetAddr &) 0x0000000170028e50: {
  m_addr = {
    _union = {
      direct = "d\U00000014\U0000001e("
      indirect_contents = (indirect = "", capacity = 0)
    }
    _size = 4
  }
  m_net = NET_IPV4
  m_scope_id = 0
}

(lldb) print address.GetLinkedIPv4()
(uint32_t) 1679040040

(lldb) print address.GetAddrBytes()
(std::vector<unsigned char>) size=16 {
  [0] = '\0'
  [1] = '\0'
  [2] = '\0'
  [3] = '\0'
  [4] = '\0'
  [5] = '\0'
  [6] = '\0'
  [7] = '\0'
  [8] = '\0'
  [9] = '\0'
  [10] = '\xff'
  [11] = '\xff'
  [12] = 'd'
  [13] = '\x14'
  [14] = '\x1e'
  [15] = '('
}

[pinheadmz]

h/t to @0xB10C and #338 which brought this issue to light!

[willcl-ark]

Dang, that's a nice bugfix fren!

Patch LGTM.

We'll want to rebuild the images after this one. I'll kick them all off in a bit on my machine, and then merge if they all complete?

[pinheadmz]

Just realized I made a mistake as well will update....

[pinheadmz]

Fixed my mistake and added an addrman unit test for reviewers

[pinheadmz]

added a test that pulls the new v27.0 image and checks addrman buckets. also had to bump lnd version: https://x.com/roasbeef/status/1781086945986404559

[willcl-ark]

ACK fa0e383

tested locally, and it works nicely