fix: reject a bool IP address in-library instead of silently corrupting the packet

[JarryShaw]

Fixes #491 — the unfixed shared root of #469/#481, which fixed the same mechanism at exactly one call site.

The defect

bool is an int subclass, and ipaddress.ip_address() treats any int below 2**32 as IPv4. So on an IPv4-typed field a bool was silently accepted and corrupted the packet — no exception, no warning:

IPv4.make(src=True, dst=False).pack()  ->  4500001400000000001100000000000100000000
decoded back                           ->  src=0.0.0.1  dst=0.0.0.0

The IPv4/IPv6 asymmetry is why #481 looked sufficient. On an IPv6-typed field the same conversion happens to raise, because the resulting IPv4Address's version mismatches — so the IPv6 half masked the IPv4 half, and fixing MH._make_opt_mn_id alone left the root untouched.

The fix

A shared _reject_bool(value, description) helper in pcapkit/corekit/fields/ipaddress.py, following the module's existing _reraise_as_field_value_error pattern, called as the first statement of _IPAddressField.pre_process, IPv4InterfaceField.pre_process and IPv6InterfaceField.pre_process — before any ipaddress.* dispatch.

That placement is deliberate and matches #481's: a correct check in the wrong position does not fire, and that exact mistake has been made twice in this repository's history with a passing test each time.

pcapkit/protocols/internet/esp.py gets the equivalent guard in SecurityAssociation.__init__, ahead of its direct ipaddress.ip_address() call, which sits outside the Field abstraction entirely and had no version check at all.

Exception choice

FieldValueError in the fields module, ProtocolError in esp.py — each module's existing convention, and ProtocolError is what #481 used for the same mechanism in a protocol module.

Deliberately not pcapkit.utilities.exceptions.BoolError: its docstring is "The argument(s) must be bool type" — it is for the inverse case, where a bool was expected and something else arrived. Using it here would invert its meaning.

Verification

Every public field class in the module, enumerated rather than spot-checked, with pcapkit.__file__ asserted against this branch before import:

IPv4AddressField.pre_process(True/False)   -> FieldValueError
IPv6AddressField.pre_process(True/False)   -> FieldValueError
IPv4InterfaceField.pre_process(True/False) -> FieldValueError
IPv6InterfaceField.pre_process(True/False) -> FieldValueError

Both original reproductions:

IPv4.make(src=True, dst=False)          -> FieldValueError: invalid IP address: must not be a bool,
                                           not True -- pass int(True) if the numeric value is what is wanted
SecurityAssociation(spi=1, destination=True) -> ProtocolError: invalid destination: must not be a bool, ...

The escape hatch the message promises actually works, and ordinary input is unaffected:

pre_process(int(True))    -> b'\x00\x00\x00\x01'
pre_process(int(False))   -> b'\x00\x00\x00\x00'
pre_process(1)            -> b'\x00\x00\x00\x01'
pre_process('10.0.0.1')   -> b'\n\x00\x00\x01'
pre_process(IPv4Address)  -> b'\n\x00\x00\x01'
interface('10.0.0.1/24')  -> b'\n\x00\x00\x01\xff\xff\xff\x00'

Revert-proof. Neutering the three field guards and the esp guard in place:

11 failed, 43 passed, 53 subtests passed
  SUBFAILED(field='IPv6InterfaceField', value=True) ...reject_a_bool_value_for_every_version
  FAILED ...test_ipv4_make_rejects_a_bool_address_through_the_public_api
  SUBFAILED(destination=True) ...test_security_association_rejects_a_bool_destination

Restored: 44 passed, 63 subtests passed.

The new tests sweep {True, False} across the address and interface types for both IP versions, because the IPv6-raises/IPv4-corrupts asymmetry is precisely what let this survive #481.

Scope deliberately left out

The roughly ten further IPv4-typed maker sites listed in #491 as "confirmed by inspection but not executed" are untouched here: RROption/LSROption/SSROption in ipv4.py, the dual-stack option makers in mh.py, and SCTP._make_param_ipv4. They share this mechanism and very likely still corrupt, but those files are being edited concurrently for #490/#493/#494, and splitting the change across them would have produced conflicting edits to the same lines.

With the root guarded, every one of those sites that routes through an IPv4AddressField is now covered automatically. SCTP._make_param_ipv4 and any site calling ipaddress.* directly are not, and want a follow-up pass once the other streams land.