const: bring the 16 bespoke registries onto EnumRegistry, after fixing the base's str-key dispatch

[JarryShaw]

Is your feature request related to a problem? Please describe.

Filed per @JarryShaw's ruling on #842: "Take (1) then and file tracking for the remaining work and keep on working."
#842 is closed as done at 111 of 127 const enum classes inheriting EnumRegistry. This tracks the remaining 16.

They divide into two groups, measured on main:

4 StrEnum classes  ftp/command.FEATCode, http/method.Method, pcapng/option_type.OptionType,
                   reg/apptype.AppType
12 others          ftp/command.CommandType (IntFlag), ftp/return_code.{ResponseKind,GroupingInformation},
                   http/status_code.StatusCode (all IntEnum), reg/apptype.TransportProtocol,
                   and the four transport subclasses dccp.DCCP / sctp.SCTP / tcp.TCP / udp.UDP

Each defines its own __new__ and its own hand-written get, so none shares pcapkit/vendor/default.py's template —
which is why #858 could not sweep them up.

The blocker is real, not merely bookkeeping. The base get dispatches on isinstance(key, str) and its str
branch never falls back to the value path. So on a StrEnum registry a valid value that is not a name would
start raising KeyError where the bespoke get resolved it. That was measured during #858's review on a real
EnumRegistry + StrEnum fixture, not hypothesised.

Describe the solution you'd like

Two steps, in order:

1. Fix the base's str-key dispatch so a StrEnum registry can resolve a key that is a valid value but not a
   name — try the name path, then fall through to the value path rather than raising. tests/const has no
   StrEnum-valued registry on the base today, so this needs a fixture as well as a fix.
2. Then convert the 16, or decide per class that its bespoke __new__ earns an exemption. ftp/return_code and
   http/status_code are IntEnum and unaffected by (1), so they could go first.

Describe alternatives you've considered

Converting them without (1), which #858's review showed regresses StrEnum value lookups. Rejected.

Leaving all 16 permanently bespoke. Defensible — they genuinely differ — but it leaves get/get_all/register/
register_alias unavailable on four public StrEnum registries including AppType, which is the one most likely to
be reached for.

Additional context

A second, independent item for the same 16, from #859's review: ftp/return_code.py:284,299 and
http/status_code.py:250,265 still carry their own -1-as-no-default convention (if default == -1: raise). After
#859 the 111 treat -1 as an ordinary value, so the repo will hold two meanings for a literal -1 default (plus
a third, value-to-mint, at pcapng/option_type.py:201 and six sites in protocols/). Harmonising that belongs here
too.

Related: #842 (closed, the 111), #858 (the template move), #859 (the NO_DEFAULT sentinel), #775 (the minting tier).

[JarryShaw]

Checkable blocker. Step 1 of this issue — fixing the base get's str-key dispatch so it falls through to the
value path — edits pcapkit/corekit/enum.py, which open PR #859 is currently rewriting (NO_DEFAULT → a
dedicated sentinel type, per the owner's ruling). Starting here now would collide in that one file.

gh pr view 859 -R JarryShaw/PyPCAPKit --json state -q .state     # must be MERGED (or CLOSED)

Until then there is nothing to start. Once #859 lands, step 1 is a small, self-contained change to
EnumRegistry.get plus a StrEnum fixture — tests/const has no StrEnum-valued registry exercising the base
today, which is why the gap went unnoticed until #858's review probed for it.

Step 2 (converting the 16, or exempting them case by case) depends on step 1 and on nothing else.

[JarryShaw]

Scope correction — this issue is larger than filed, and it gates 78% of #775's conversion work.

I filed this as "bring the 16 bespoke registries onto EnumRegistry", framed as method availability. That
undersold it. Verified by AST walk over every _missing_ on main:

minting extend_enum branches, whole tree : 1007
of which on these 16 classes             :  781  (78%) across 6 files / 9 classes
  reg/apptype/apptype.py  AppType      766
  http/status_code.py     StatusCode     8
  ftp/return_code.py      ResponseKind / GroupingInformation / ReturnCode   3
  ftp/command.py          FEATCode / Command                                2
  http/method.py          Method         1
  pcapng/option_type.py   OptionType     1

AppType alone holds 766 of them.

Why that matters here rather than on #775: #775's ruling converts placeholder-minting branches to
_unregistered_member. These 9 classes do not inherit EnumRegistry, so they have no _unregistered_member to
convert to. #775's classification confirms almost all 781 are placeholder mints that should convert — AppType's
766 are 754 unassigned_%d plus 4 reserved_%d plus 8 genuinely-named IANA services — but none of it can be done
until these classes have the base.

So this issue is the gate for that work, and the ordering is: #859 → this → #775's remaining 781 branches. The
step-1 str-key dispatch fix matters more than I first wrote, too: four of the nine are StrEnum
(AppType, FEATCode, Command, Method, OptionType), so they are exactly the population the base get's
missing value-path fallback would break.

The blocker above is unchanged — gh pr view 859 --json state must read MERGED, since step 1 edits
pcapkit/corekit/enum.py which #859 is rewriting.

[JarryShaw]

Unblocked — #859 merged at 21:04:58Z and it was the gate. The condition on the blocker above now reads clear:

gh pr view 859 --json state -> MERGED
pcapkit/corekit/enum.py on main: class NoDefaultType at :58, NO_DEFAULT = NoDefaultType() at :219
                                 the str-key branch this issue must change is at :291

So pcapkit/corekit/enum.py is no longer being rewritten by an open PR, and step 1 is a self-contained change to one
branch of one method.

A reminder of what step 1 actually is, since it is easy to under-scope. EnumRegistry.get's str path at
enum.py:291 does cls._member_map_[key] and, on KeyError, either raises or falls back to default — it never
tries the value path. On a StrEnum registry that means a key which is a valid value but not a name raises
KeyError where the bespoke get resolved it. Four of the sixteen are StrEnum — ftp/command.FEATCode,
http/method.Method, pcapng/option_type.OptionType, reg/apptype.AppType — so they are exactly the population the
gap would break.

tests/const has no StrEnum-valued registry exercising the base today, which is why the gap went unnoticed until
#858's review probed for it with a synthetic fixture. So step 1 needs a real fixture as well as the fix.

And the scope correction from earlier still stands: these 16 classes hold 781 of the 1007 minting _missing_
branches — AppType alone 766 — so this issue also gates that part of #775. #861 is converting the other 226 now.

blocked removed, wip applied.