Fix epoch CPU admission for hot-add

[Alan Jowett (Alan-Jowett)]

Description

Fixes #4732.

This updates epoch CPU tracking to support systems where the maximum logical processor count exceeds the active processor count at startup.

• size the epoch backing table for the maximum logical processor count while admitting only active CPUs into the consensus ring
• use processor-change callbacks to initialize and admit hot-added CPUs in two phases
• fail fast if current-CPU epoch APIs are reached before a CPU finishes admission
• add unit coverage for startup active-CPU subsets, hot-add admission, and the unadmitted-CPU fail-fast case
• document the active/admitted CPU model and hot-add behavior

Testing

Not run per request.

If new tests were added:

• Unit tests are added.
• Driver tests are added.
• Fuzz tests are added.

Documentation

Updated docs/EpochBasedMemoryManagement.md.

Installation

No installer impact.

[Copilot]

Pull request overview

This PR updates the runtime epoch-based memory reclamation subsystem to tolerate systems where the kernel-reported maximum logical CPU count exceeds the active CPU count at startup, and to support CPU hot-add by admitting CPUs into the epoch participant set in phases.

Changes:

• Reworks ebpf_epoch CPU tracking to allocate per-CPU backing state for the maximum CPU count while only admitting active CPUs into the consensus/participant ring (with hot-add admission via processor-change callbacks).
• Adds unit tests to cover startup active-CPU subsets, hot-add admission behavior, and an unadmitted-CPU fail-fast path (child-process based).
• Adjusts libfuzzer harness/core lifecycle management and updates CI test filters and epoch documentation accordingly.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/libfuzzer/include/fuzz_helper_function.hpp	Makes ebpf core init/term optional per helper instance and hardens cleanup paths.
tests/libfuzzer/core_helper_fuzzer/libfuzz_harness.cpp	Centralizes ebpf core lifetime in the fuzz harness and disables per-helper core lifecycle management.
libs/runtime/unit/platform_unit_test.cpp	Adds unit tests for startup active subset, hot-add admission, and fail-fast on unadmitted CPU (child process).
libs/runtime/ebpf_epoch.c	Implements admitted-CPU ring + processor change callback path for initializing/admitting CPUs and routing epoch consensus only across admitted CPUs.
docs/EpochBasedMemoryManagement.md	Documents the maximum-vs-admitted CPU model and hot-add admission behavior.
.github/workflows/cicd.yml	Excludes the new fail-fast epoch tests from multiple CI jobs.
.azure/OneBranch.PullRequest.yml	Excludes the new fail-fast epoch tests from multiple Azure pipeline jobs.

[Alan Jowett (Alan-Jowett)]

Anurag Saxena (@saxena-anurag) You are 100% correct. I missed that the hot-add complete occurs after the CPU can be scheduled. That totally breaks my current assumption. Revisiting this.