Normalize Linux audio pipeline to lossless Float32 (skill: lossless-audio-compliance)

- native bridge WAV writers: 16-bit PCM truncation -> IEEE Float32 (fmt 3, bit-exact)
- vendor sheredom_json.h (vst3sdk 3.8.1 lacks moduleinfo/json.h)
- Dockerfile: drop nonexistent sfizz tag v1.2.3
- Python sf.write -> subtype=FLOAT everywhere
- resample_poly -> kaiser(16) window (>=130dB stopband)
- audio_editor._write_wav_lossless(): FLOAT passthrough, TPDF dither only for
  8/16/24-bit final exports
- app.jsx WAV encoders: Float32 default, TPDF dither for integer depths, fix 24-bit
- tests: lossless compliance gate (null test < -140dBFS, float32 format) 9/9 pass
- TASKS_WINDOWS_AUDIO_NORMALIZE.md: Windows-side remaining tasks

native_bridge/src/RenderJob.cpp

@@ -9,9 +9,17 @@
 #include "RenderJob.h"
 #include "NativeInstrumentEngine.h"
 
-// sheredom/json.h (public domain, vendored with the VST3 SDK) is header-only
-// and needed for SF2/SFZ render even without the VST3 SDK — include it directly.
+// sheredom/json.h (public domain) is header-only and needed for SF2/SFZ render
+// even without the VST3 SDK. Vendored copy: native_bridge/include/sheredom_json.h
+// (the pinned VST3 SDK v3.8.1 has no moduleinfo/json.h); fall back to the SDK
+// copy when a newer checkout provides it.
+#if __has_include("sheredom_json.h")
+#include "sheredom_json.h"
+#elif __has_include("vst3sdk/public.sdk/source/vst/moduleinfo/json.h")
 #include "vst3sdk/public.sdk/source/vst/moduleinfo/json.h"
+#else
+#error "sheredom/json.h not found — vendored copy at native_bridge/include/sheredom_json.h"
+#endif
 #ifdef HAVE_VST3SDK
 #include "public.sdk/source/vst/vstpresetfile.h"
 #include "public.sdk/source/common/memorystream.h"
@@ -75,7 +83,10 @@ int64_t memberInt(const json_object_s* o, const char* key, int64_t def) {
     return memberNumber(o, key, d) ? (int64_t)d : def;
 }
 
-// --- WAV writer (stdlib only, 16-bit PCM stereo, little-endian) --------------
+// --- WAV writer (stdlib only, IEEE Float32 stereo, little-endian) ------------
+// Lossless: keeps the full float32 render buffer bit-exact. No PCM truncation
+// and no dither here — dither/quantization belongs ONLY to a final export
+// stage (lossless-audio-compliance skill, Rules 4 & 7). Format tag 3.
 void writeU16(std::ofstream& f, uint16_t v) {
     char b[2] = { (char)(v & 0xFF), (char)((v >> 8) & 0xFF) };
     f.write(b, 2);
@@ -91,12 +102,12 @@ void writeWavHeader(std::ofstream& f, uint32_t sampleRate) {
     f.write("WAVE", 4);
     f.write("fmt ", 4);
     writeU32(f, 16);          // fmt chunk size
-    writeU16(f, 1);           // PCM
+    writeU16(f, 3);           // IEEE float
     writeU16(f, 2);           // stereo
     writeU32(f, sampleRate);
-    writeU32(f, sampleRate * 4);  // byte rate
-    writeU16(f, 4);               // block align
-    writeU16(f, 16);              // bits per sample
+    writeU32(f, sampleRate * 8);  // byte rate = sr * 2ch * 4B
+    writeU16(f, 8);               // block align
+    writeU16(f, 32);              // bits per sample
     f.write("data", 4);
     writeU32(f, 0);           // patched in finishWav
 }
@@ -108,14 +119,10 @@ void finishWav(std::ofstream& f, uint64_t dataBytes) {
     f.flush();
 }
 void writeFrames(std::ofstream& f, const float* L, const float* R, uint32_t n) {
+    // IEEE float32 little-endian, bit-exact: no clamp, no quantization.
     for (uint32_t i = 0; i < n; ++i) {
-        auto cl = [](float v) -> int {
-            if (v > 1.0f) v = 1.0f;
-            else if (v < -1.0f) v = -1.0f;
-            return (int)(v * 32767.0f);
-        };
-        writeU16(f, (uint16_t)cl(L[i]));
-        writeU16(f, (uint16_t)cl(R[i]));
+        f.write(reinterpret_cast<const char*>(&L[i]), 4);
+        f.write(reinterpret_cast<const char*>(&R[i]), 4);
     }
 }