test: correct docstrings on the #572 truncated-option tests

[JarryShaw]

Summary

Follow-up to #578, which merged before its cross-review finished. The
independent review (on a different model, per this repo's practice for
agent-raised changes) found the two new tests — TCPUDPUnitTests and
IPv4UnitTests' test_a_truncated_option_still_parses_its_declared_length
— correct in every constructed byte, assertion, and pass/fail outcome, but
flagged several factual errors in their docstring prose:

• 0x4f is TCP's Option.Reserved_79, not an "unassigned" kind (the
  UnassignedOption schema class handles it, but the wire code itself is
  reserved, not unassigned).
• The IPv4 docstring attributed the option_padding rewind-and-reread-as-
  padding mechanism to "OptionField.unpack never returns for a well-formed HOPOPT header with an SMF_DPD option #431 machinery" inside OptionField.unpack. It is
  actually in Schema.unpack (pcapkit/protocols/schema/schema.py:890),
  added by Fix seven PCAP-NG parser defects (#341-#347) #371, and predates OptionField.unpack never returns for a well-formed HOPOPT header with an SMF_DPD option #431 — OptionField.unpack never returns for a well-formed HOPOPT header with an SMF_DPD option #431's own contribution to
  OptionField.unpack is only the post-break progress check, which performs
  no rewind in this case.
• The IPv4 docstring claimed declaring an 8-octet option area would trip
  IPv4's stricter length-sum check "before the accommodation under test is
  ever reached." That's mechanically false: the accommodation does run (the
  short data field is read and left-padded), the outer
  _read_ipv4_options check just discards that result afterwards, once its
  loop over the parsed options finishes.
• The TCP docstring attributed the "sizes by what it consumed, not the
  declared length" measurement to OptionField.unpack; it is actually
  TCP._read_tcp_options itself (pcapkit/protocols/transport/tcp.py:698,
  counter += len(schema)).
• The TCP docstring's opening sentence ("cut short mid-option") was wrong for
  the TCP fixture specifically — nothing in it is truncated (hdr_len == len(raw), and the test asserts a full round-trip); the over-declaration
  is internal to the option, not the capture. Reworded. Left the IPv4
  opening line as-is, since that fixture genuinely is short (28 physical vs
  36 declared octets).
• The TCP docstring's justification for checking length=32 alongside 12
  ("the fix would reject both identically") argued for one case being
  sufficient rather than explaining why both are checked; replaced with the
  actual distinction — the pad width scales with the declared length (24
  zero octets vs 4).
• Switched both tests' fixed 6-octet trailing literal from bytes([0xaa, ...]) to bytes.fromhex('aabbccddeeff'), matching the surrounding files'
  idiom.

No assertion, constructed byte, or test outcome changes here — this is a
docstring-only correction. Both tests still pass, and both still fail
against a reject-on-any-shortfall guard in FieldBase.unpack with the
FieldValueError text quoted in #572.

Test plan

• pytest tests/protocols/transport/test_tcp_udp_unit.py tests/protocols/internet/test_ipv4_unit.py -k test_a_truncated_option_still_parses_its_declared_length -q — 2 passed, 2 subtests passed, unchanged from before this correction
• Full unit tier on this branch: pytest -q --ignore=tests/integration --ignore-glob='*_runtime.py' --ignore-glob='*_regression.py'