OpenClaw Signals: Memory Refactors, Preload Guards, and Execution Tuning

PR

Refactor: Isolate Memory Reclaim into Interface and Module

Moves memory reclamation logic into its own subsystem, decoupling it from procedural housekeeping in the lifecycle manager. If you are hitting accumulation issues in long-running processes, this isolation is the path to actually fixing it without spinning cycles on unrelated logic.

Fix: Do Not Show Failed Descendants in All Tasks

Consolidates filtering logic for task expansion to ensure failed tasks don't continue to clutter the active view just because their parents are running. It's a small UX win, but crucial for teams trying to debug workflow states without visual noise.

Fix: Skip Preload if Provisioning is Disabled

Patches the preload behavior so it doesn't error out if you have disabled configuration provisioning entirely. This unblocks minimal setups that don't need the config lifecycle overhead.

Fix: Properly Close Pipeline Indexes After Optimization

Addresses a resource management issue where pipeline indexes weren't being closed out after optimization jobs ran. If you are dealing with memory pressure after heavy build cycles, this patch stops the leak.

Issue

Feature Discussion: Attaching Nodes to Flow Branches

A feature request to let conversation branches ride along with flow executions, not just simple tasks. This is a solid thread if you are trying to integrate complex, branching LLM logic directly into the orchestration layer.