Improve the CPU pinning logic for BatchExecutor.

[dmkozh]

Description

• De-prioritize core 0 to decrease the probability of clashing with OS house-keeping
• Pin hyper-threaded cores first if we run out of physical cores, and never wrap around the list (leave the over-subscribed rest workers unpinned)

Checklist

• Reviewed the contributing document
• Rebased on top of master (no merge commits)
• Ran clang-format v8.0.0 (via make format or the Visual Studio extension)
• Compiles
• Ran all tests
• If change impacts performance, include supporting evidence per the performance document

[Copilot]

Pull request overview

Improves Linux CPU affinity ordering for BatchExecutor.

Changes:

• De-prioritizes CPU 0’s physical core.
• Uses hyper-thread siblings after distinct physical cores.
• Leaves excess workers unpinned.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/util/BatchExecutor.h	Updates pinning state and documentation.
src/util/BatchExecutor.cpp	Implements topology-aware CPU ordering and oversubscription handling.

Comments suppressed due to low confidence (1)

src/util/BatchExecutor.cpp:94

• This comment repeats the inaccurate absolute-last guarantee: on restricted or heterogeneous affinity sets, later sibling tiers can follow CPU 0. The rotation only guarantees that CPU 0 follows the other allowed CPUs from its own core.

        // Demote CPU 0 behind its own hyper-thread siblings so it becomes the
        // very last CPU we pin.

[graydon]

Nice! Thanks.