fix(hipdnn): handle default streams in kpack launches (ALMIOPEN-2506)

[BrianHarrisonAMD]

Summary

Fixes default-stream handling along the public hipDNN backend -> kernel-ingestor -> packed-kernel execution path found during review of #11392. This draft targets that PR's branch so the fix can be incorporated before it merges.

JIRA ID: ALMIOPEN-2506 (ticket)

Risk Assessment

2 / 5 - narrow stream-classification changes. Null, legacy, and per-thread defaults resolve through the live current device during preparation and bypass concrete-stream ownership queries at launch. Concrete-stream validation, module-device binding/restoration, and the existing per-device properties cache remain intact. No compute kernels, public APIs, or CMake behavior change.

ASIC Coverage

Verified on gfx942 (two MI300X GPUs, then one visible GPU) and gfx950 (one MI355X). This is focused packed-pointwise coverage, not every engine, graph shape, or Windows configuration.

Testing Summary

Built the real backend/provider and packed fixtures with the ingestor and its existing rocKE dependency enabled.

Configuration	Passed	Expected cross-device skips
gfx942, two GPUs	87	0
gfx942, one visible GPU	82	5
gfx950, one GPU	82	5

All three public default-stream cases selected the packed engine and matched the CPU reference. Ordinary-stream and broken-archive controls passed. SLURM jobs: 67931487 and 67931512.

Testing Checklist

• Scoped pre-commit checks on all eight PR files.
• Backend device/logging tests (hipdnn_backend_tests, scoped filters).
• Resolver and launch suites (hip_kernel_provider_tests; *TestHandleDeviceResolver* and *TestProgram*:*TestKernel*).
• Public packed-graph suite (hip_kernel_provider_integration_tests --gtest_filter='*IntegrationGpuKernelIngestorKpack*').
• PR CI for the updated head.

Technical Changes

• Recognize default tokens before device-ownership queries in backend logging, backend heuristics, the ingestor resolver, and packed launches.
• Add live-device resolver and public packed-graph regressions; retain launch output/restoration and rejection coverage.

Allocation bounds inside archive extraction and path-replacement races remain outside this fix; post-extraction checks and observed-link rejection do not eliminate those limitations.

[codecov]

Codecov Report

❌ Patch coverage is 93.33333% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ugin_sdk/include/hipdnn_plugin_sdk/DeviceQuery.hpp	91.67%	1 Missing ⚠️

Additional details and impacted files

@@                                   Coverage Diff                                   @@
##           users/ggilbert/ALMIOPEN-2506/kpack-runtime-followup   #12015      +/-   ##
=======================================================================================
- Coverage                                                87.02%   87.01%   -0.02%     
=======================================================================================
  Files                                                      456      457       +1     
  Lines                                                    61683    61711      +28     
  Branches                                                  9084     9088       +4     
=======================================================================================
+ Hits                                                     53678    53693      +15     
- Misses                                                    5072     5082      +10     
- Partials                                                  2933     2936       +3     

Flag	Coverage Δ
hipDNN	87.01% <93.33%> (-0.02%)	⬇️

Files with missing lines	Coverage Δ
...hipdnn/backend/src/heuristics/DeviceProperties.cpp	84.21% <100.00%> (+0.43%)	⬆️
...hipdnn/backend/src/heuristics/DeviceProperties.hpp	100.00% <ø> (ø)
projects/hipdnn/backend/src/logging/Logging.cpp	92.19% <100.00%> (ø)
...ugin_sdk/include/hipdnn_plugin_sdk/DeviceQuery.hpp	62.96% <91.67%> (ø)

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.