Seven register/issubclass guards in the protocol classes still leak a bare TypeError for a non-class #1026

All seven register classmethods in pcapkit/protocols/ carry a bare if not issubclass(x, ProtocolBase): with no isinstance(x, type) ahead of it, so a non-class argument is rejected by issubclass before the guard's own raise is reached and the caller gets TypeError: issubclass() arg 1 must be a class leaked out of abc instead of the library's own exception.

Measured on main, raise site read off traceback.extract_tb(...)[-1] rather than off the message:

Transport.register is included, and an earlier version of this body wrongly excluded it. Its UnsupportedCall at transport.py:110 is gated on if cls is Transport:, so it fires only for a literal call on the abstract class — which is never how it is reached. The guard below it leaks through the concrete subclasses:

Transport.register  -> UnsupportedCall | transport.py:110
TCP.register        -> TypeError       | <frozen abc>:123
UDP.register        -> TypeError       | <frozen abc>:123

grep -rn "if not issubclass" pcapkit/ finds 13 sites: six are fixed in #1025, and these seven remain.

The per-site question

The maintainer's ruling on #1021 was "code follows prose": add an explicit class check so callers get the registry's own exception. That ruling was scoped to the register_* module functions, and these are a different surface — register classmethods on protocol classes, keyed by code.

Measured: all seven Raises: clauses already promise RegistryError for "not a … subclass", and none says "class" explicitly. So the fix is a correction rather than a contract addition, on the reading that an int is not a subclass of anything — but the wording is ambiguous enough that each clause wants rewording to "is not a class, or not a … subclass" alongside the code change.

Two inconsistencies worth settling in the same pass, both pre-existing: the Raises: prose alternates between promising a Protocol subclass and a ProtocolBase subclass across these seven sites, while every runtime message says "Protocol subclass" and every guard checks ProtocolBase. That is the same conflation #514 is about, so the guard targets should not move here — only the prose should stop contradicting itself.

Not breaking when fixed: RegistryError subclasses TypeError through BaseError, so except TypeError keeps working and only code matching the exact type or the old message text sees a difference.

Worth doing as one change rather than seven, since the fix is mechanical and the tests are parallel. Related: #1021, #1025, #514.