Add MicroCUDA backend, adaptive output wait, wider native K coverage (improve perf)

Shim
- Adaptive output-event wait: sleep to ~120 us before the predicted
  completion (smoothed per wait site), spin near it, back off when late.
  D4R_SHIM_OUTPUT_POLL_US restores a fixed interval.

- Defer the input-marker wait into the worker thread with VRAM interop
  (D4R_SHIM_DEFER_MARKER, default on), taking the host setup off the
  critical path.

MicroCUDA
- Experimental ZLUDA-free backend in wine_nvcuda_bridge, plus
  tools/microcuda and scripts/build_microcuda.sh. Serves accuracy, fast
  and native kernels; unregistered modules fall back to the reference
  path. Unclassified writes invalidate prepared weights.

M kernels
- SWIN_PREP_IDENTITY_AT: key weight prep by allocation identity
  (MicroCUDA); enc3 tube uses it with 8 slots. ZLUDA ignores it.

- dec2 (gfx1101): SWIN_PACKED_RESIDUAL, byte-identical, ~1.6% faster.

- Opt-in gfx12 hardware fp8 conversion (SWIN_FP8_HWCVT), unverified
  on hardware.

Native K translators
- ctile: handle variants that re-materialize the parameter base (mvlo).

- rewrite_output: check live-in registers, not just outputs.

- build.sh: native input covers all 8 flag variants, output the 4 HDR
  "max" variants; depth-tile skip stays mvhi-only.

Tools
- dump_runner and bench_dlss_kernel harness additions.

tools/microcuda/arch_spoof.c

@@ -0,0 +1,35 @@
+/* LD_PRELOAD for zluda_precompile.py: report D4R_SPOOF_GFX (e.g. gfx1201) as the device's gcnArchName, so
+ * ZLUDA compiles (and caches) code objects for a GPU that is not installed. Loading them then fails, which
+ * the precompile treats as expected; the cache entry is written before the load. Never use it for running. */
+#define _GNU_SOURCE
+#define __HIP_PLATFORM_AMD__
+#include <hip/hip_runtime_api.h>
+#include <dlfcn.h>
+#include <stdlib.h>
+#include <string.h>
+
+hipError_t hipGetDevicePropertiesR0600(hipDeviceProp_tR0600* props, int device)
+{
+    static hipError_t (*real)(hipDeviceProp_tR0600*, int);
+    if (!real)
+        real = (hipError_t (*)(hipDeviceProp_tR0600*, int))dlsym(RTLD_NEXT, "hipGetDevicePropertiesR0600");
+    if (!real)
+    {
+        /* ZLUDA loaded RTLD_LOCAL (ctypes): its HIP is not in the global scope RTLD_NEXT searches */
+        void* hip = dlopen("libamdhip64.so.7", RTLD_NOW | RTLD_NOLOAD);
+        if (!hip)
+            hip = dlopen("libamdhip64.so", RTLD_NOW | RTLD_NOLOAD);
+        if (hip)
+            real = (hipError_t (*)(hipDeviceProp_tR0600*, int))dlsym(hip, "hipGetDevicePropertiesR0600");
+    }
+    if (!real)
+        return hipErrorNotInitialized;
+    hipError_t result = real(props, device);
+    const char* arch = getenv("D4R_SPOOF_GFX");
+    if (result == hipSuccess && props && arch && *arch)
+    {
+        memset(props->gcnArchName, 0, sizeof(props->gcnArchName));
+        strncpy(props->gcnArchName, arch, sizeof(props->gcnArchName) - 1);
+    }
+    return result;
+}