fix(FIR): size q15 state buffer for ARM_MATH_LOOPUNROLL, not just ARM_MATH_DSP

arm_fir_init_q15() only allocates the larger (numTaps+blockSize)
state buffer when ARM_MATH_DSP is defined, otherwise it allocates the
smaller (numTaps+blockSize-1).

arm_fir_q15()'s non-MVE processing function has a separate
ARM_MATH_LOOPUNROLL-gated code path (4-sample-at-a-time, using
read_q15x2_ia()/__SMLALD()/__PKHBT(), which all have portable
non-hardware-DSP fallback implementations and so do not themselves
require ARM_MATH_DSP) that reads up to index numTaps+blockSize-1 from
the state buffer - one element past what's allocated when
ARM_MATH_LOOPUNROLL is defined without ARM_MATH_DSP, an out-of-bounds
read.

Verified with a standalone host reproduction of the exact
read_q15x2_ia/__SMLALD/__PKHBT access pattern, placing the state
buffer immediately before a PAGE_NOACCESS guard page: with the current
(numTaps+blockSize-1)-sized buffer the reproduction crashes with an
access violation; sized to (numTaps+blockSize) it completes cleanly.

Fix: allocate the larger buffer whenever ARM_MATH_DSP *or*
ARM_MATH_LOOPUNROLL is defined, matching what arm_fir_q15()'s
loop-unrolled path actually needs regardless of whether hardware DSP
instructions or their software fallback are in use.

Fixes #305

Signed-off-by: 94xhn <87560781+94xhn@users.noreply.github.com>

Author: 94xhn

Source/FilteringFunctions/arm_fir_init_q15.c

@@ -75,7 +75,7 @@
       {0.3, -0.3, 0, 0}.
   </pre>
                    <code>pState</code> points to the array of state variables.
-                   <code>pState</code> is of length <code>numTaps+blockSize</code>, when ARM_MATH_DSP is defined. Otherwise, it is of length <code>numTaps+blockSize-1</code> where <code>blockSize</code> is the number of input samples processed by each call to <code>arm_fir_q15()</code>.
+                   <code>pState</code> is of length <code>numTaps+blockSize</code>, when ARM_MATH_DSP or ARM_MATH_LOOPUNROLL is defined (arm_fir_q15()'s loop-unrolled path reads up to <code>numTaps+blockSize-1</code>, requiring this larger size regardless of whether ARM_MATH_DSP itself is defined). Otherwise, it is of length <code>numTaps+blockSize-1</code> where <code>blockSize</code> is the number of input samples processed by each call to <code>arm_fir_q15()</code>.
  
   @par          Initialization of Helium version
                    For Helium version the array of coefficients must be a multiple of 8 (8a) even if less
@@ -95,7 +95,7 @@ ARM_DSP_ATTRIBUTE arm_status arm_fir_init_q15(
 {
   arm_status status;
 
-#if defined (ARM_MATH_DSP)
+#if defined (ARM_MATH_DSP) || defined (ARM_MATH_LOOPUNROLL)
 
   /* The Number of filter coefficients in the filter must be even and at least 4 */
   if (numTaps & 0x1U)
@@ -139,7 +139,7 @@ ARM_DSP_ATTRIBUTE arm_status arm_fir_init_q15(
 
   return (status);
 
-#endif /* #if defined (ARM_MATH_DSP) */
+#endif /* #if defined (ARM_MATH_DSP) || defined (ARM_MATH_LOOPUNROLL) */
 
 }