llama.cpp: build the unified llama binary

[julien-c]

Enable -DLLAMA_BUILD_APP=ON to install the unified llama binary introduced in ggml-org/llama.cpp#23296. The new binary dispatches to serve / cli subcommands; existing llama-completion, llama-server, etc. continue to be installed alongside.

Tag is bumped to b9265 because b9260 predates #23296.

Smoke test verifies the new binary is on PATH and prints its command list (it exits 1 when invoked with no args).

---

• Have you followed the guidelines for contributing?
• Have you ensured that your commits follow the commit style guide?
• Have you checked that there aren't other open pull requests for the same formula update/change?
• Have you built your formula locally with HOMEBREW_NO_INSTALL_FROM_API=1 brew install --build-from-source <formula>?
• Is your test running fine brew test <formula>?
• Does your build pass brew audit --strict <formula> (after doing HOMEBREW_NO_INSTALL_FROM_API=1 brew install --build-from-source <formula>)? If this is a new formula, does it pass brew audit --new <formula>?

---

• AI was used to generate or assist with generating this PR.

Claude Code drafted the formula diff (cmake flag, tag bump, smoke test) under my direction. I verified that -DLLAMA_BUILD_APP=ON produces a working build/bin/llama by building llama.cpp@b9265 locally with the same flag outside the formula; I have not yet run brew install --build-from-source llama.cpp against the formula itself. brew style and brew audit (non-strict) pass.

[julien-c]

@carlocab @cho-m i am proposing for your consideration that we opt-in to building the new llama binary entry point that was recently merged in ggml-org/llama.cpp#23296

cc @botantony for your thoughts too

Thanks!

[carlocab]

Does upstream use this configuration in the binaries they build?

[julien-c]

good question, let's see where ggml-org/llama.cpp#23488 lands maybe

[julien-c]

ggml-org/llama.cpp#23462

[carlocab]

I see it's on by default, but that's not even in a release yet. Let's wait for it to bake a bit before turning it on here. It'll likely be enabled on the next autobump PR anyway.