Split the vld1q_s8 + vzip1q_u64/vzip2q_u64 4x16c8 NEON I8MM GEMM/IGEMM microkernels into a separate _zip variant (VARIANT=ZIP) and update the existing c8-neoni8mm microkernels to use vld2q_dup_u64 for lane 0 instead of zero-initializing a loop-carried uint64x2x2_t register pair for vld2q_lane_u64(..., 0). - #11794

…EMM/IGEMM microkernels into a separate `_zip` variant (`VARIANT=ZIP`) and update the existing `c8-neoni8mm` microkernels to use `vld2q_dup_u64` for lane 0 instead of zero-initializing a loop-carried `uint64x2x2_t` register pair for `vld2q_lane_u64(..., 0)`.

Benchmark on `qs8_qc8w_gemm_minmax_fp32_ukernel_4x16c8__neoni8mm` (MobileNetV2 `M=49, N=1280, K=320`, GOPS):

```text
uarch                   lane (GOPS)     dup (GOPS)     zip (GOPS)
-------------------  --------------  -------------  -------------
Neoverse-V2 (Grace)           373.2          277.3          406.6
Neoverse-V2 (Axion)           362.4          263.9          390.1
Neoverse-N3                   252.1          233.5          274.1
AmpereOne                     133.4           98.5          134.3
Apple M6                      392.5          371.8          454.3
Apple M5                      377.2          342.3          435.7
Oryon-L                       544.5          527.4          560.8
Oryon-M                       268.0          269.0          288.6
Oryon V3-L                    464.2          482.9          501.1
Oryon V3-M                    250.4          253.8          262.5
Cortex-X4                     302.1          300.0          305.3
Cortex-X3                     199.6          197.1          197.5
Cortex-X2                     269.1          256.7          278.9
Cortex-C1-Ultra               249.0          248.5          249.2
Cortex-C1-Pro                 179.9          175.7          175.2
Cortex-A725                   208.9          205.1          205.5
Cortex-A720                   226.8          236.3          236.2
Cortex-A715                   161.1          162.5          162.9
Cortex-A710                   159.5          158.7          166.3
Cortex-A520                    58.2           84.5           97.8
Cortex-A510r1                  55.2           67.7           72.5
Cortex-A510                    54.9           58.1           56.0
```

```bash
bazel build -c opt //bench:qs8_qc8w_gemm_fp32_bench
bazel-bin/bench/qs8_qc8w_gemm_fp32_bench \
  --benchmark_filter="qs8_qc8w_gemm_minmax_fp32_ukernel_4x16c8__neoni8mm" \
  --benchmark_min_time=0.5s
```

PiperOrigin-RevId: 1005800878