fix: fall back to AVX2 f16 kernels on AVX-512 CPUs

[paramt]

Fix f16 SIMD dispatch on x86 CPUs that have AVX-512F but do not have native AVX512-FP16 support.

Today, those CPUs are classified as SimdSupport::Avx512. The f16 dispatch for L2, dot, cosine, and norm L2 handles Avx512FP16 and Avx2, but not plain Avx512, so it falls through to the scalar implementation. This change routes plain Avx512 to the AVX2 f16 kernels while keeping Avx512FP16 on the native AVX512-FP16 kernels.

This matches the existing bf16 behavior:

• L2:

  lance/rust/lance-linalg/src/distance/l2.rs

  Lines 183 to 185 in 5266629

  	SimdSupport::Avx2 | SimdSupport::Avx512 => unsafe {
  	bf16_kernel::l2_bf16_avx2(x.as_ptr(), y.as_ptr(), x.len() as u32)
  	},

• Dot:

  lance/rust/lance-linalg/src/distance/dot.rs

  Lines 153 to 155 in 5266629

  	SimdSupport::Avx2 | SimdSupport::Avx512 => unsafe {
  	bf16_kernel::dot_bf16_avx2(x.as_ptr(), y.as_ptr(), x.len() as u32)
  	},

• Cosine:

  lance/rust/lance-linalg/src/distance/cosine.rs

  Lines 119 to 121 in 5266629

  	SimdSupport::Avx2 | SimdSupport::Avx512 => unsafe {
  	bf16_kernel::cosine_bf16_avx2(x.as_ptr(), x_norm, y.as_ptr(), y.len() as u32)
  	},

• Norm L2:

  lance/rust/lance-linalg/src/distance/norm_l2.rs

  Lines 118 to 120 in 5266629

  	SimdSupport::Avx2 | SimdSupport::Avx512 => unsafe {
  	bf16_kernel::norm_l2_bf16_avx2(vector.as_ptr(), vector.len() as u32)
  	},

[westonpace]

Makes sense to me. Thanks! Will merge when CI passes

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!