header_rewrite: %{CIDR:,N} zeroes the IPv4 mask instead of keeping the default

[zwoop]

ConditionCidr::set_qualifier() parses the whole qualifier with a single strtol, so an empty field and an explicit 0 are indistinguishable. The documented v6-only form therefore masks IPv4 away entirely.

Docs (doc/admin-guide/plugins/header_rewrite.en.rst):

%{CIDR}         Defaults to 24,48 (as above)
%{CIDR:16}      IPv4 CIDR mask is 16 bits, IPv6 mask is 48
cond %{CIDR:,8} ="fd00::" #note the IPv6 Mask is in the second position

For %{CIDR:,8}, strtol() consumes no digits, returns 0 with endp at the comma, 0 <= 32 passes, and _v4_cidr becomes 0 rather than staying at its default of 24. Masking a client address through the cidr.h helpers:

%{CIDR:,8}    (v4=0, today)     client 10.2.3.4 -> 0.0.0.0
%{CIDR:24,8}  (v4=24, docs)     client 10.2.3.4 -> 10.2.3.0

%{CIDR:24,} has the same problem on the other side (v6 becomes 0 instead of 48), and %{CIDR:abc} silently means v4=0 instead of raising a TSError.

Suggested fix in set_qualifier() — treat "no digits consumed" as "keep the default", and only allow it when the field really was empty:

  cidr = strtol(q.c_str(), &endp, 10);
  if (endp == q.c_str()) {
    // No v4 digits: only the empty-field form "%{CIDR:,N}" is valid; keep the default.
    ok   = (*endp == ',' || *endp == '/' || *endp == ':');
    cidr = _v4_cidr;
  }

plus the same treatment after the separator. That also turns %{CIDR:abc} into an error rather than a silent /0.

Noticed while reviewing #13205, which pads the missing field with the documented default (24) — that padding is correct against the docs, but disagrees with what the plugin does today. Related: tools/hrw4u validates cidr() as range(1, 32) / range(1, 128) while the plugin accepts 0 in both positions, so an explicit %{CIDR:0,0} cannot be expressed in hrw4u at all.