Prebuilt: pin the penalties sampler fix (#95)

[danielhanchen]

Pins c26185b from #95 into the nightly prebuild set.

#95 stops llama_sampler_penalties_apply from probing the frequency map once per candidate. It now walks token_count (at most penalty_last_n entries) and indexes cur_p directly, with a guard that falls back to the old scan when an earlier sampler has reordered or dropped candidates.

This matters for the prebuilds because our own Studio presets enable it: the qwen3.5 and qwen3.6 family profiles set presence_penalty: 1.5, and on a 248320 token vocab the old scan cost about 11% of generation throughput.

Qwen3.6-27B-UD-Q4_K_XL, 400 tokens, temperature 0, median of 4, throughput from llama.cpp's own /metrics:

presence_penalty	build	server t/s
1.5	stock	130.29
1.5	patched	146.66

Output is unchanged: 16 stock vs patched comparisons across 4 prompts, both penalty settings, greedy and temperature 0.8 with a fixed seed, all byte identical. tests/test-sampling passes.

Per the notes in pr-set.json the entry stays listed until the change lands upstream, since merging it into fork master alone would drop it from the mix.