[mono][jit] Optimize Vector128.Create on Arm64

[matouskozak]

When initializing Vector128 with constant float/double value (e.g., Vector128.Create(2.0f)) the values are loaded into 32/64-bit register (fmov) and then duplicated into 128-bit register (dup.4s/2d). For immediate floating-point constants on Arm64 this can be optimized into a single fmov FMOV (vector, immediate) which copies the immediate constant into every element of the 128-bit register.

With AOT LLVM the behavior is already optimized and it uses fmov.4s/2d to directly fill the 128-bit register.

Steps to take:

1. Move the Arm64 immediate floating-point constant check functionality to a separate function so that it can be used both for float/double types as well as for vector types.

   runtime/src/mono/mono/mini/mini-arm64.c

   Lines 4876 to 4906 in ff2de36

   	// Arm64 floating-point modified immediate constant check (2 * 128 combinations)
   	// Float: aBbbbbbc defgh000 00000000 00000000
   	// Double: aBbbbbbb bbcdefgh 00000000 00000000 00000000 00000000 00000000 00000000
   	// Trailing zeros check
   	if ((r_imm & mask_constant) == 0) {
   	// Mask for b
   	guint8 mask_b;
   	int idx_last_b;
   	if (is_double) {
   	mask_b = 0xFF;
   	idx_last_b = 54;
   	} else {
   	mask_b = 0x1F;
   	idx_last_b = 25;
   	}
   	guint8 masked_b = (r_imm & ((guint64)mask_b << idx_last_b)) >> idx_last_b;
   	int size = is_double ? 64 : 32;
   	// NOT(B) == b check
   	if (((r_imm & ((guint64)1 << (size - 2))) && masked_b == 0)
   	|| (!(r_imm & ((guint64)1 << (size - 2))) && masked_b == mask_b)) {
   	//imm8 = abcdefgh
   	guint8 imm8 = ((r_imm & ((guint64)1 << (size - 1))) >> (size - 8))
   	| ((r_imm & ((guint64)1 << idx_last_b)) >> (idx_last_b - 6))
   	| ((r_imm & ((guint64)0x3F << (idx_last_b - 6))) >> (idx_last_b - 6));
   	arm_fmov_imm(code, (is_double ? 0x01 : 0x00), imm8, dreg);
   	break;
   	}
   	}

2. Use ^ check inside vector constructor code path to emit single fmov for immediate constants.