Add Infinity_Vis_1 performance core

Infinity_Vis_1/ARCHITECTURE.md

@@ -0,0 +1,44 @@
+# Infinity_Vis_1 Architecture
+
+## Design goal
+
+Make the render path cheaper than the current application by replacing
+object-heavy frame construction with a compact, controller-oriented runtime.
+
+## High-level layers
+
+1. `mapping_xml.py`
+   Parses and writes the existing XML mapping format.
+2. `models.py`
+   Small `slots=True` dataclasses for tiles, segments, controllers, scenes, and frames.
+3. `patterns.py`
+   Stateless pattern functions that return tile colors only.
+4. `renderer.py`
+   Owns precomputed controller routes and reusable `bytearray` payloads.
+5. `engine.py`
+   Separates output cadence from preview cadence and handles delta sending.
+6. `output/ddp.py`
+   Sends already-built payloads as DDP packets.
+7. `presets.py`
+   JSON preset persistence.
+8. `benchmark.py`
+   Simple measurable entry point for render throughput.
+
+## Main performance choices
+
+- Render tile colors first, not rich nested preview objects.
+- Reuse controller payload buffers every frame.
+- Precompute segment byte offsets once at startup.
+- Keep the runtime pure-stdlib for the new core.
+- Keep preview as a consumer of a compact frame snapshot, not as the center of the engine.
+- Decouple preview timing from hardware output timing so UI cost cannot directly cap live output.
+- Skip unchanged controller payloads while still sending periodic keepalives.
+
+## What is intentionally not in phase 1
+
+- Full Qt UI parity
+- Full preview painter parity
+- WLED discovery UI port
+- Advanced transition compositing parity
+
+Those belong in phase 2, after the faster runtime is proven and benchmarked.