Fix segfault on startup when built with AVX (-march=native etc.)

The FastMaths SIMD helpers tell the compiler the buffers they receive are
32-byte aligned via __builtin_assume_aligned(ptr, 32). The buffers actually
passed in are not guaranteed to be: the SoundEngine buffers come from plain
`new float[]` (16-byte aligned on x86-64), SoundEngine::addData runs
addVectors() on the sound card output buffer the engine does not allocate,
and multiply4()/multiply8() receive sinc-table coefficients at arbitrary
offsets (&sincWindow[frac * order + j]).

When polyphone is compiled with AVX enabled (any -march that turns on AVX,
e.g. -march=native), GCC trusts the 32-byte promise and auto-vectorizes the
generic loops with aligned 256-bit moves (vmovaps %ymm). Whenever a buffer is
not actually 32-byte aligned at runtime, that move raises a general-protection
fault and the audio thread crashes on startup. Distro/baseline builds
(-march=x86-64, SSE only, 16-byte movaps) never hit this.

Remove the unsafe alignment assumptions. Auto-vectorization still applies with
unaligned moves (no measurable cost on current x86), and the aarch64 NEON
paths already use unaligned vld1q_f32.

Upstream: https://github.com/davy7125/polyphone/pull/331

--- a/core/fastmaths.cpp
+++ b/core/fastmaths.cpp
@@ -86,8 +86,6 @@

 void FastMaths::addVectors(float * __restrict a, const float * __restrict b, unsigned int size)
 {
-    a = (float*)__builtin_assume_aligned(a, 32);
-    b = (const float*)__builtin_assume_aligned(b, 32);
     unsigned int i = 0;

 #ifdef __aarch64__
@@ -113,7 +111,6 @@

 void FastMaths::clamp(float * values, unsigned int size)
 {
-    values = (float*)__builtin_assume_aligned(values, 32);
     unsigned int i = 0;

 #ifdef __aarch64__
@@ -141,8 +138,6 @@

 void FastMaths::multiplyAdd(float * __restrict data, const float * __restrict dataToMultiplyAndAdd, unsigned int size, float coeff)
 {
-    data = (float*)__builtin_assume_aligned(data, 32);
-    dataToMultiplyAndAdd = (float*)__builtin_assume_aligned(dataToMultiplyAndAdd, 32);
     unsigned int i = 0;

 #ifdef __aarch64__
@@ -169,8 +164,6 @@

 void FastMaths::multiply(float * __restrict data, const float * __restrict dataToMultiply, unsigned int size, float coeff)
 {
-    data = (float*)__builtin_assume_aligned(data, 32);
-    dataToMultiply = (const float*)__builtin_assume_aligned(dataToMultiply, 32);
     unsigned int i = 0;

 #ifdef __aarch64__
@@ -196,8 +189,6 @@

 float FastMaths::multiply8(const float * __restrict coeffs, const qint16 * __restrict srcData16, const quint8 * __restrict srcData24)
 {
-    coeffs = (const float*)__builtin_assume_aligned(coeffs, 32);
-
 #ifdef __aarch64__
     // Load 8 int16 and convert to int32
     int16x8_t s16 = vld1q_s16(srcData16);
@@ -241,9 +232,6 @@

 float FastMaths::multiply4(const float * __restrict coeffs, const float * __restrict srcDataF)
 {
-    coeffs = (const float*)__builtin_assume_aligned(coeffs, 32);
-    srcDataF = (const float*)__builtin_assume_aligned(srcDataF, 32);
-
 #ifdef __aarch64__
     // Load all values
     float32x4_t a = vld1q_f32(coeffs);