test(rccl): add host microtests for the GIN host proxy

[aduforest]

JIRA ID: AICOMRCCL-2807

What

Host-only microtests for src/gin/gin_host_proxy.cc and the versioned GPU-context writers in src/gin/proxy_gpucontext/ (gpucontext{,_v1,_v2}.cc), in a new rccl-UnitTestsMicroGinHostProxy binary (64 tests).
Overlaps rahulvaidya20's #13221 (same file name, same unit); will reconcile with it.

How

• gin-host-proxy-test.cc #includes the four hipified units and seams the RMA plugin, device allocations, hipMemset and the DMA-BUF export in the TU; GFDs are posted straight into the host-side queues.
• Pins 4 production bugs as observed, not fixed:
  • ncclGinProxyRegMrSym: when both DMA-BUF registrations fail, it returns before closing the second exported fd.
  • ncclGinProxyCreateContext: accepts backendVersion 0, fails in the GPU-context writer, and releases nothing it created.
  • A non-power-of-two maxRecvs gives a queue the index mask cannot wrap.
  • ncclGinProxyDestroyContext never frees the signal-offsets buffer.

Coverage

Lines: gin_host_proxy.cc 476/476, gpucontext.cc 33/33, gpucontext_v1.cc 10/10, gpucontext_v2.cc 13/13.

Mutation Testing

57/57 killed: 53 in gin_host_proxy.cc, 4 in the GPU-context writers.

Test Plan

Standalone build: rccl-UnitTestsMicroGinHostProxy 64/64; --gtest_shuffle --gtest_repeat=20 is clean.

[aduforest]

@argus review

[therock-pr-bot]

✅ All Policy Checks Passed

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

🎉 All policy checks passed!

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

[amd-argus]

ARGUS

Findings

5 more finding(s), not posted inline

• projects/rccl/test/host/gin-host-proxy-test.cc:858 (Low) ginCtx_ left dangling after mid-test destroyContext; a failing Create then makes TearDown destroy it twice
• projects/rccl/test/host/gin-host-proxy-test.cc:49 (Low) Outer for loops without braces around a nested if or a wrapped body (lines 49, 1114, 1295)
• projects/rccl/test/host/gin-host-proxy-test.cc:906 (Low) Create_LogsWhenGdrCopyIsUnavailable proves its no-GDR half only through log text
• projects/rccl/test/host/gin-host-proxy-test.cc:808 (Nit) WriteCombined assertion pins a host_flags argument that gdrwrap.h allocMemCPUAccessible ignores
• projects/rccl/test/host/gin-host-proxy-test.cc:989 (Nit) Pinned-bug tests use '// Observed:' instead of the repo's 'LATENT BUG (file:line)' convention

This PR changes 1,552 lines across 7 files. A change this size is hard to review
well, for a person and for this bot alike: reviewers lose the thread, defects hide in the
volume, and the feedback arrives too late to be cheap to act on. Splitting it into a
stacked pull request
gives each piece a reviewable diff and keeps the history. This review covers the whole change
regardless.

Bottom line: Looks approval ready.

_{Comment @argus review to re-run, or @argus re-review after pushing.}

[aduforest]

Addressed in 53517f8 (all five, rebuilt, 64/64 and clean × 20 shuffled):

• :858 dangling ginCtx_: every mid-test destroy now goes through a Destroy() helper that nulls ginCtx_, so a failing Create can't make TearDown destroy twice.
• Loop braces: added on the three nested/wrapped loops.
• GDR half: Create_GdrCopyChoosesTheCountersMappingTheGpuSees now also checks the counters pointer published in the GPU context (host pointer without GDRCopy, the device mapping with it), not just the log line.
• WriteCombined: dropped the host_flags assertions; only forceHost (which the HIP path honours) is still checked.
• Pinned bugs now use LATENT BUG (gin_host_proxy.cc:line).

@argus review

[amd-argus]

ARGUS

This PR changes 1,558 lines across 7 files. A change this size is hard to review
well, for a person and for this bot alike: reviewers lose the thread, defects hide in the
volume, and the feedback arrives too late to be cheap to act on. Splitting it into a
stacked pull request
gives each piece a reviewable diff and keeps the history. This review covers the whole change
regardless.

Bottom line: LGTM.

_{Comment @argus review to re-run, or @argus re-review after pushing.}

[aduforest]

@argus review

[amd-argus]

ARGUS

This PR changes 1,558 lines across 7 files. A change this size is hard to review
well, for a person and for this bot alike: reviewers lose the thread, defects hide in the
volume, and the feedback arrives too late to be cheap to act on. Splitting it into a
stacked pull request
gives each piece a reviewable diff and keeps the history. This review covers the whole change
regardless.

Bottom line: LGTM.

_{Comment @argus review to re-run, or @argus re-review after pushing.}