metal: implement DSv4 Lightning Indexer

[tarruda]

Overview

Metal backend for GGML_OP_LIGHTNING_INDEXER

Additional information

This is the first of a series of changes that target improving DSv4 performance on Apple Silicon. I extracted this commit from my dsv4-improvements branch, which includes metal backends for hyper-connections and a few other improvements.

On master branch, these are some performance numbers of DSv4 on my M1 Ultra (llama-bench --mmap 1, -fa 1, -p 512, -n 128; d=0/10k/20k/30k, benchmarking this quant):

• pp512: 153.73 ± 0.87 t/s
• tg128: 8.91 ± 0.04 t/s
• pp512 @ d10000: 73.90 ± 0.39 t/s
• tg128 @ d10000: 8.66 ± 0.03 t/s
• pp512 @ d20000: 45.83 ± 0.18 t/s
• tg128 @ d20000: 8.26 ± 0.03 t/s
• pp512 @ d30000: 33.40 ± 0.21 t/s
• tg128 @ d30000: 7.94 ± 0.01 t/s

With this commit, things improve slightly:

• pp512: 155.19 ± 0.91 t/s
• tg128: 8.95 ± 0.04 t/s
• pp512 @ d10000: 86.95 ± 0.69 t/s
• tg128 @ d10000: 9.00 ± 0.05 t/s
• pp512 @ d20000: 62.01 ± 0.45 t/s
• tg128 @ d20000: 8.68 ± 0.04 t/s
• pp512 @ d30000: 49.18 ± 0.33 t/s
• tg128 @ d30000: 8.60 ± 0.02 t/s

With all commits from my branch, this is what it looks like:

• pp512: 168.67 ± 0.82 t/s
• tg128: 23.59 ± 0.06 t/s
• pp512 @ d10000: 151.10 ± 0.55 t/s
• tg128 @ d10000: 20.87 ± 0.13 t/s
• pp512 @ d20000: 145.98 ± 0.45 t/s
• tg128 @ d20000: 20.68 ± 0.13 t/s
• pp512 @ d30000: 141.91 ± 0.29 t/s
• tg128 @ d30000: 20.56 ± 0.06 t/s

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, this is 100% generated by GPT 5.6 Sol. I have reviewed and tested the changes, but I am not a Metal kernel developer and thus cannot judge the quality of these changes. Hopefully the maintainers can extract something as the compounded performance improvements can make DSv4 flash very usable with a local coding harness. CC @ggerganov @fairydreaming