AtomLane v0.17.0 — Adaptive Host-Wide Scheduling

AtomLane 0.17.0 adds adaptive host-wide scheduling for atomic_exec workloads.

Highlights:

• Coordinates multiple concurrent AtomLane sessions with a cross-process host capacity ledger.
• Adapts to Apple heterogeneous cores, SMT systems, homogeneous CPUs, live CPU pressure, memory, power and thermal state.
• Evaluates a whole atom batch, applies fair-share admission, and lets uncontended throughput work borrow safe idle capacity.
• Refreshes capacity during long runs and exposes active host sessions and global occupied slots in live progress.
• Recovers stale reservations safely and fails closed on malformed or unsafe scheduler state.

Verification:

• 331 tests in the published source-bound report, 0 failures, all 9 release checks passed.
• GitHub CI passed on macOS and Windows with Python 3.10–3.13, plus CodeQL, plugin and secret scans.
• Controlled macOS benchmark: 4 independent 310-second tasks, 3.9973× measured speedup, 929.98 seconds saved.
• Controlled Windows benchmark: 4 independent 310-second tasks, 3.9945× measured speedup, 929.58 seconds saved.

Benchmark results measure the controlled independent workload and are not a universal speedup claim. The host-wide scheduler currently applies to atomic_exec; compatibility parallel_exec behavior is unchanged.

Install or upgrade instructions and the live report are linked from the repository README.