[bri-prism]

What

Docs only. Updates the upstream llama.cpp support claims for Bonsai 2.

• The community benchmark README and the llama.cpp template still said the Q2_0 file "currently requires our fork" and linked the old dev repo copy. Both now link Ternary-Bonsai-2-27B-Q2_0.gguf in the main repo and say it runs on upstream llama.cpp b11556 or newer.
• README, AGENTS.md, BACKEND-SUPPORT.md and MODEL-FORMATS.md said "current official llama.cpp". They now name release b11556.
• The upstream support section now lists all four sizes. It also says the MTP and DFlash2 drafters on the model cards need our fork.

Why

People checking whether Bonsai 2 runs on stock llama.cpp were reading the stale "requires our fork" text. A release number gives them a concrete check.

How it was checked

Upstream master and release b11556 were tested against the prism fork on the 2B, 4B, 9B and 27B Q2_0 files:

• Mean KLD is about 0 and the top token matches 99.9% of the time, on Metal and CPU.
• Greedy text and image output are identical.
• The commit before b11556 gives garbage (perplexity above 1 million).
• Upstream rejects PTQ1_0 and PQ2_0 as unknown types.
• Upstream loads the DFlash2 drafters but accepts none of their drafts.

[khosravipasha]

Thanks LGTM