chore(agentx): bump AIPerf to agentx-v1.0.3 [no breaking changes, no perf effect, no need to rebase]

[cquil11]

Summary

• Bump the utils/aiperf submodule from agentx-v1.0.2 (743fb82ab798dc898fce98d0321db61e7564942a) to agentx-v1.0.3 (bcfc235c552a17de6d1a7a5d2345cf012401588d).
• Centralize the released AIPerf pin on main instead of carrying equivalent source changes independently in multiple configuration PRs.
• Release: https://github.com/SemiAnalysisAI/aiperf/releases/tag/agentx-v1.0.3
• Upstream fix: fix(dataset): make mmap conversation reads concurrency-safe agentx-harness#37

Compatibility and performance

• No breaking changes.
• No performance-related changes.
• The upstream change only makes mmap-backed conversation reads concurrency-safe by replacing shared-cursor seek()/read() operations with position-independent byte-range slicing.
• It does not change AgentX request payloads, workload pacing, concurrency, synthetic AL, latency timing, metric calculations, routing, or inference-server execution.

Existing PRs

There is no immediate or strict necessity to rebase existing PRs solely to continue testing this AIPerf change. PRs #2516 and #2517 currently pin the pre-merge commit 67a42758b366589951cdecc06087e0a6cf88603d, while the agentx-v1.0.3 release tag points to bcfc235c552a17de6d1a7a5d2345cf012401588d. Those commits have the identical AIPerf tree (4beb3f27c46efaa66ba960b7ea2c0df2c6a8cd04) and therefore identical source and behavior.

Because the gitlink SHAs differ, after this PR lands, #2516 and #2517 may need a minimal one-line gitlink alignment to the released commit before merge if Git reports a submodule conflict. That alignment does not require a new performance sweep or a substantive rebase. No existing PR branch was rebased or modified while preparing this PR.

Validation

• Verified agentx-v1.0.3 resolves to bcfc235c552a17de6d1a7a5d2345cf012401588d.
• Verified 67a42758b366589951cdecc06087e0a6cf88603d and bcfc235c552a17de6d1a7a5d2345cf012401588d have the identical AIPerf tree.
• Verified the InferenceX diff changes only utils/aiperf.
• Verified the release is public, non-draft, non-prerelease, and marked Latest.

[claude]

Claude finished @cquil11's task in 1m 41s —— View job

---

LGTM - no blocking issues found

Verified the diff changes only the utils/aiperf gitlink (743fb82 → bcfc235c552a17de6d1a7a5d2345cf012401588d, matching the SHA stated in the PR description), and .gitmodules still points at the public SemiAnalysisAI/aiperf repo. No master configs or benchmark/launch scripts touched. Note: I could not independently resolve the agentx-v1.0.3 tag upstream from this environment (network access not permitted in this run), so the tag→SHA mapping relies on the PR description.

[claude]

LGTM — straightforward submodule pin bump.

What was reviewed: confirmed the diff is a single gitlink change to utils/aiperf (743fb82 → bcfc235), pinning to the published agentx-v1.0.3 release; no other files in this repo are touched, so there's no InferenceX-side logic to review beyond the pointer update.

Extended reasoning...

Overview

The entire diff is a one-line gitlink update to the utils/aiperf submodule, moving the pinned commit from 743fb82 (agentx-v1.0.2) to bcfc235 (agentx-v1.0.3). No other files in the InferenceX repo are modified.

Security risks

None. This repo's diff contains no executable code changes — only a pointer to a different upstream commit of a separate, already-vetted repository. The PR description characterizes the upstream change as a concurrency-safety fix (byte-range slicing instead of shared-cursor seek/read for mmap-backed reads), which is not security-sensitive.

Level of scrutiny

Minimal scrutiny is appropriate. Submodule bumps like this are mechanical, and the description gives a specific, verifiable target commit and release tag along with a clear rationale (centralizing the pin instead of duplicating it across config PRs). This is the same category of change as a version bump or dependency pin update.

Other factors

This is a benchmarking-tooling dependency (AIPerf) rather than production inference-serving code, further lowering blast radius if the upstream change had subtle issues. No prior review history exists on this PR.