rccl/gin: fix inverted system-scope fence guard on HIP backends

[dlamd1dai]

Summary

• The GIN Put/PutValue device templates gate the pre-put system fence on (required==system && given>required). On HIP, hip_compat.h orders thread_scope_system as the max value, so given>required is never true when required==system — the __threadfence_system() branch is dead code, and a caller that only guaranteed a weaker scope (e.g. block) silently skips the system fence it requested.
• Flip the guard to (required==system && given<required) on the two HIP backends (anvil_sdma, rocshmem_gda), routed through a new NCCL_GIN_THREADFENCE_SYSTEM() seam in gin_device_common.h (expands to __threadfence_system() in production; unit tests can override it to observe the branch).
• gdaki intentionally left as given>required: it is CUDA-only (NCCL_GIN_GDAKI_ENABLE=0 on HIP) and libcu++ orders thread_scope_system as the lowest value — the opposite of HIP — so given>required is correct there, and flipping it would make its DOCA system release dead code. Its comment is expanded to document this.
• Tests: install a counting override of the fence seam and replace the former Put_WeakerGivenScopeStillPuts with Put_WeakerGivenScopeFencesAndPuts (required=system, given=block → asserts the fence ran exactly once and the put landed), plus Put_EqualScopeTakesNoFence (required=given=system → asserts no fence and the put landed).

Follow-up to review on #2 (stacked on that branch).

JIRA ID: AICOMRCCL-1527

Stacking

Branch	Tip	Notes
Base	#2 39a05ff	GIN device-template unit tests + CI wiring
This PR	3b66d26	2 commits, 5 files (+75/−19 vs #2)

Merge order: ROCm#10420 → #1 → #2 → this PR → #4.

Test plan

• Rebased onto rccl: GIN AllToAll device-template unit tests (gin-gda + gin-sdma shadow fix) #2 (39a05ff); branch now stacks correctly (was previously a sibling off meta-repro).
• Fence-counting GDA template tests: Put_WeakerGivenScopeFencesAndPuts and Put_EqualScopeTakesNoFence.
• Prior MI300X (gfx942) fixture run: 34/34 PASS (13 GDA + 12 SDMA + 9 IPC); log pr3-fixtures-full-20260822-001621.log.
• MI355 functional validation sufficient; not chasing fork CI timeouts.
• gdaki path unchanged (CUDA-only; not built/tested in this AMD/HIP tree).

Made with Cursor

[dlamd1dai]

Superseded by upstream PR ROCm/rocm-systems#10675, stacked on #10672. Branch users/dondai/gin-scope-fence-guard-fix rebased onto users/dondai/gin-gda-a2a-unittests @ 11fb115 (2 commits, tip a2899ae). Post-rebase MI300X validation: 34/34 PASS (log ~/rocm-systems/ddai-artifacts/logs/pr3-fixtures-full-20260824-231854.log). Closing this fork PR.