[dart2wasm] Expose missing Wasm SIMD intrinsics (bitmask, signed comparisons, shuffle, shifts) in dart:_wasm #63680

In #63521 and https://dart-review.googlesource.com/c/sdk/+/507760, we lowered core arithmetic and basic comparison operations on Float32x4, Int32x4, and Float64x2 in simd.dart to hardware WebAssembly vector primitives (WasmF32x4, WasmI32x4, WasmF64x2). This produced a 3.19x speedup on real-world vector kernels.

However, several standard operations in dart:typed_data still fall back to scalar lane extraction loops or temporary heap memory buffering because the underlying WebAssembly v128 instructions are not yet exposed in dart:_wasm (wasm_types.dart) or the dart2wasm backend.

Missing Intrinsics & Optimization Opportunities

1. Vector Sign Bitmask Extraction (i32x4.bitmask)

• Missing Wasm Instruction: i32x4.bitmask
• Current Fallback: Int32x4.signMask and Float32x4.signMask currently extract sign bits by pulling 4 separate integer lanes (extractLane(i)) and bit-shifting them into a 4-bit scalar mask.
• Proposed Solution: Expose bitmask on WasmI32x4 in wasm_types.dart.
• Impact: Collapses signMask evaluation from multi-instruction lane extractions down to 1 single CPU hardware instruction.

2. Signed Integer Vector Comparisons (i32x4.lt_s, gt_s, le_s, ge_s)

• Missing Wasm Instructions: i32x4.lt_s, i32x4.gt_s, i32x4.le_s, i32x4.ge_s
• Current Fallback: Int32x4.greaterThan(), lessThan(), greaterThanOrEqual(), and lessThanOrEqual() currently execute via scalar lane extraction and comparison loops because WasmI32x4 only implements equality (eq).
• Proposed Solution: Add signed integer comparison intrinsics to WasmI32x4 in wasm_types.dart and map them in compiler code generators (pkg/dart2wasm/).
• Impact: Allows integer SIMD comparison logic to execute 100% in vector registers without scalar unpacking.

3. Hardware Vector Lane Shuffle (i8x16.shuffle)

• Missing Wasm Instruction: i8x16.shuffle (for general vector swizzling)
• Current Fallback: Float32x4.shuffle(), Float32x4.shuffleMix(), Int32x4.shuffle(), and Int32x4.shuffleMix() currently use temporary Int32List memory buffers and array indexing (_list[mask & 0x3]).
• Proposed Solution: Expose i8x16.shuffle in WasmI32x4 and WasmF32x4.
• Impact: Eliminates all heap memory buffer roundtrips during vector lane permutations.

4. Vector Integer Bitwise Shifts (i32x4.shl, shr_s, shr_u)

• Missing Wasm Instructions: i32x4.shl, i32x4.shr_s, i32x4.shr_u
• Current Status: Missing from WasmI32x4 interface.
• Proposed Solution: Add parallel bitwise shift methods to WasmI32x4.
• Impact: Essential for hardware acceleration of packed color manipulation (ARGB/RGBA), cryptography, and integer SIMD algorithms.