net: change CNetAddr::ip to have flexible size

Before this change `CNetAddr::ip` was a fixed-size array of 16 bytes,
not being able to store larger addresses (e.g. TORv3) and encoded
smaller ones as 16-byte IPv6 addresses.

Change its type to `prevector`, so that it can hold larger addresses and
do not disguise non-IPv6 addresses as IPv6. So the IPv4 address
`1.2.3.4` is now encoded as `01020304` instead of
`00000000000000000000FFFF01020304`.

Rename `CNetAddr::ip` to `CNetAddr::m_addr` because it is not an "IP" or
"IP address" (TOR addresses are not IP addresses).

In order to preserve backward compatibility with serialization (where
e.g. `1.2.3.4` is serialized as `00000000000000000000FFFF01020304`)
introduce `CNetAddr` dedicated legacy serialize/unserialize methods.

Adjust `CSubNet` accordingly. Still use `CSubNet::netmask[]` of fixed 16
bytes, but use the first 4 for IPv4 (not the last 4). Only allow
subnetting for IPv4 and IPv6.

Co-authored-by: Carl Dong <contact@carldong.me>

src/netbase.cpp

@@ -13,6 +13,7 @@
 
 #include <atomic>
 #include <cstdint>
+#include <limits>
 
 #ifndef WIN32
 #include <fcntl.h>
@@ -838,8 +839,8 @@ bool LookupSubNet(const std::string& strSubnet, CSubNet& ret)
         if (slash != strSubnet.npos)
         {
             std::string strNetmask = strSubnet.substr(slash + 1);
-            int32_t n;
-            if (ParseInt32(strNetmask, &n)) {
+            uint8_t n;
+            if (ParseUInt8(strNetmask, &n)) {
                 // If valid number, assume CIDR variable-length subnet masking
                 ret = CSubNet(network, n);
                 return ret.IsValid();