Initial commit: testpattern → fakesink pipeline over MXL shared memory

- NodeBase, Signal, FlowDef, V210 shared headers
- testpattern node: SMPTE 75% color bars writer at 25fps
- fakesink node: non-blocking MXL reader with per-second stats
- studio-manager: FlowGraph data model, graph-driven fork/exec launcher
- mxl pinned as submodule at 0ae1dc5

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -0,0 +1,48 @@
+# DMF Studio — AI Context
+
+Read ARCHITECTURE.md first for the full picture. This file adds guidance specific to working in this codebase.
+
+## Project in one sentence
+
+Node-based media signal processing system: each node is a separate C++ process, nodes exchange video/audio via MXL shared memory, studio-manager orchestrates them, Vue.js frontend planned.
+
+## Current state
+
+- Core built: testpattern node (color bars writer), fakesink node (reader/logger), studio-manager (fork/exec orchestrator)
+- Not built yet: real processing nodes, Vue frontend, graph.json persistence, WebSocket API
+
+## Key conventions
+
+- Node configuration comes from `NODE_CONFIG` env var (JSON). Never hardcode flow UUIDs.
+- MXL domain comes from `MXL_DOMAIN` env var. Default: `/dev/shm/dmf-studio` (Linux) or `/tmp/dmf-studio` (macOS).
+- All shared utilities live in `shared/` as header-only files. New nodes add a subdirectory under `nodes/`.
+- Timing always goes through MXL time API (`mxlGetCurrentIndex`, `mxlGetNsUntilIndex`, `mxlSleepForNs`). Do not use `std::chrono::system_clock` for frame timing — MXL uses TAI.
+- V210 line stride comes from `configInfo.discrete.sliceSizes[0]` after `mxlCreateFlowWriter`. Never hardcode 5120.
+
+## What the user cares about
+
+- Dmitry is not a very experienced C++ developer. Prefer architectural discussion before implementation.
+- Keep code minimal and clear. No premature abstractions.
+- When adding a new node type, follow the patterns in `nodes/testpattern/main.cpp` and `nodes/fakesink/main.cpp`.
+
+## MXL API quick reference
+
+```
+mxlCreateInstance(domain, nullptr)
+mxlCreateFlowWriter(inst, flow_def_json, nullptr, &writer, &configInfo, &created)
+mxlCreateFlowReader(inst, flow_id, nullptr, &reader)
+mxlFlowWriterOpenGrain(writer, index, &grain, &buf)
+mxlFlowWriterCommitGrain(writer, &grain)
+mxlFlowReaderGetGrain(reader, index, timeout_ns, &grain, &buf)
+mxlGetCurrentIndex(&rate)
+mxlGetNsUntilIndex(next_index, &rate)  → ns to sleep
+mxlSleepForNs(ns)
+mxlIsFlowActive(inst, flow_id, &active)
+mxlFlowSynchronizationGroup — for multi-input nodes (PiP)
+```
+
+## MXL error codes to handle in readers
+
+- `MXL_STATUS_OK` — grain ready
+- `MXL_ERR_TIMEOUT` — writer stalled, log and retry same index
+- `MXL_ERR_OUT_OF_RANGE_TOO_LATE` — fell behind ring buffer, call `mxlGetCurrentIndex` and jump