Add batched publishing to pubsub (#281)

* Make pubsub system buffer-based to reduce syscall overhead in the hot-loop

* Remove service shutdown timeout, add debug logging for in-flight info in MetricsAggregator service

* Add python logging configuration to subscriber services

* Remove msync temporarily for ARM bottleneck

* PR comments, documentation cleanup and fixes, remove stray files

* Fix load generator flow graph, rename saturation to warmup

* Move file structure section up, remove mentioning of removed features

* Add non-first chunks to load generator data flow diagram

Author: nv-alicheng

.gitignore

@@ -196,3 +196,4 @@ examples/03_BenchmarkComparison/vllm_venv/
 # User-specific local rules (local Docker dev); do not commit
 .cursor/rules/local-docker-dev.mdc
 CLAUDE.local.md
+.cursor_artifacts/

AGENTS.md

@@ -351,6 +351,8 @@ Known failure modes when AI tools generate code for this project. Reference thes
 - **Importing removed or renamed modules**: After refactors, AI (working from stale context) may import old module paths. Always verify imports resolve to actual files.
 - **Over-documenting**: AI generates verbose docstrings, inline comments explaining obvious code, and type annotations on trivial variables. This project prefers minimal comments — only where the _why_ isn't obvious from the code.
 - **Adding backwards-compatibility shims**: If something was renamed or removed, AI may add re-exports, aliases, or deprecation wrappers. In this project, just delete the old thing and update all call sites.
+- **Empty except blocks**: Every `except` block must contain either a comment explaining why the exception is ignored, or a logging statement. Bare `except: pass` without explanation is disallowed. AI often generates empty handlers — always add the reason.
+- **No lazy imports**: All imports must be at the top of the file. Imports inside functions, methods, or conditional blocks (other than `TYPE_CHECKING`) are disallowed. The only exceptions are: (1) circular import avoidance with a documenting comment, (2) optional dependencies with a top-level try/except that sets the import to `None`, (3) security sandboxing code that intentionally restricts imports.
 
 ### Dependency & Environment