Use the libspeexdsp resampler

AFAICT it runs a little slower? but it's fixed point, and has much
better understood audio characteristics.

lib/speexdsp/win32/config.h

@@ -0,0 +1,31 @@
+// Microsoft version of 'inline'
+#define inline __inline
+
+// In Visual Studio, _M_IX86_FP=1 means /arch:SSE was used, likewise
+// _M_IX86_FP=2 means /arch:SSE2 was used.
+// Also, enable both USE_SSE and USE_SSE2 if we're compiling for x86-64
+#if _M_IX86_FP >= 1 || defined(_M_X64)
+#define USE_SSE
+#endif
+
+#if _M_IX86_FP >= 2 || defined(_M_X64)
+#define USE_SSE2
+#endif
+
+// Visual Studio support alloca(), but it always align variables to 16-bit
+// boundary, while SSE need 128-bit alignment. So we disable alloca() when
+// SSE is enabled.
+#ifndef USE_SSE
+#  define USE_ALLOCA
+#endif
+
+/* Default to floating point */
+#ifndef FIXED_POINT
+#  define FLOATING_POINT
+#  define USE_SMALLFT
+#else
+#  define USE_KISS_FFT
+#endif
+
+/* We don't support visibility on Win32 */
+#define EXPORT