DKR-R v1.0.5 Beta 15 Playtest 5 - boss victory scene correction

HISTORICAL BUILD RECORD
The original findings and limits below belong to the named earlier build.
For current instructions, use docs/README.md and CHANGELOG.md in the
source repository. These old results do not qualify later packages.


PURPOSE
The host's diagnostic playtest identified a stale scene-ID limit in the
Experimental Rollback constructor. Tricky's victory successfully marked the
reward/save and queued retail cutscene 57 (hex 39). The scene handoff then
failed with unowned-confirmed-scene-constructor because its cap was 53.
The retail level-header table has 65 entries: valid scene IDs are 0..64.
This rebuild corrects that specific rejection. Actual boss/cutscene gameplay
verification remains with the playtesters; it is not claimed fully qualified.

WHAT CHANGED
Full-game experimental scene admission now accepts the complete reviewed
retail level-ID range, including boss cinematics and endings. Non-full-scene
closures retain their original 0..53 limit. Negative/out-of-range IDs and
all other constructor ownership, service and argument checks still reject.
The two private constructor/fixture helpers use the same shared boundary.
Compile-time assertions cover the first/last IDs, boss cutscene 57 and invalid
IDs. No retail cutscene/reward sequence is skipped or rewritten.

Bounded boss-finish logging records entry/exit, the original music, reward,
transition, lead-player and save-marking calls, plus a short audio tail.
If an owned safety gate rejects an operation in this window, the log records
the reason, guest address/source, register values and guest call stack.
Milestones are flushed to runtime.log before continuing. Tracing is capped
at 160 milestone lines per finish and 300 logical tick changes afterward;
rollback duplicates also consume that cap. Unavailable raw fields appear
as 4294967295; the valid mask identifies the available global fields.
The extended window covers the observed six-second post-victory transition.

UNCHANGED
Retail instructions, registers, control flow, music, rewards, EEPROM,
cutscenes, boss AI, topology, transport, prediction and synchronization are
unchanged apart from the scene-ID admission boundary described above.
No second barrier, new signal handler, separate launcher,
test-track redirect or graphics-quality reduction has been added.
Diagnostics are outside checkpointed state and never influence gameplay.
All current Playtest 4 functionality and its explicit deferrals are retained.
The accepted Playtest 4 packages remain the rollback point.

YOUR PLAYTEST
Every participant must install this exact Playtest 5 build; do not mix
experimental packages. Use the normal launcher Backend dropdown and your
normal 2P Adventure session. No diagnostic switch needs to be enabled.
Attempt the previously failing boss finish. If practical, also check losing,
winning again, the following reward/cutscene and returning to the hub.
If it fails, keep the newest runtime.log and runtime-previous.log from BOTH
host and client before repeatedly restarting (startup rotates the log).
Use About/Support's existing log export or Open Logs location action.
For default Windows profiles: %APPDATA%\DKRPort\logs\runtime.log.
Include the boss name, win/loss and whether the host or client failed first.
Logs may contain connection information; share them privately for diagnosis.

VERIFICATION SCOPE
Compilation, source invariance, version/contract and package-content checks
only. No gameplay, self-test or online runtime qualification was performed,
following your instruction to leave all gameplay testing to you.
