chore(fish-speech): drop the darwin/metal build target

[localai-bot]

Why

The fish-speech metal-darwin-arm64 backend build has failed on every recent release (v4.5.3, v4.5.4, v4.5.5) and is a standing red on the darwin backend matrix.

fish-speech pulls tokenizers transitively from its upstream source (pip install -e fish-speech-src). On darwin/arm64 there is no prebuilt wheel for the pinned old tokenizers version, so pip builds it from source, and modern rustc rejects that old crate as a hard error:

error: casting `&T` to `&mut T` is undefined behavior, even if the reference is unused, consider instead using an `UnsafeCell`
   --> tokenizers-lib/src/models/bpe/trainer.rs:517:47
   = note: `#[deny(invalid_reference_casting)]` on by default
error: could not compile `tokenizers` (lib) due to 1 previous error

This is deterministic, not a flake. There is no clean fix that does not either pin a stale Rust toolchain or downgrade a soundness lint that guards real UB. Until upstream fish-speech moves to a tokenizers version that compiles on current toolchains, the pragmatic call is to drop the darwin target so the release backend build stays green.

Per the CLAUDE.md backend-OS-coverage guidance ("...or say in the PR why an OS is unsupported"), this is the documented reason darwin is unsupported for fish-speech.

What changed

• Removed the -metal-darwin-arm64-fish-speech entry from includeDarwin in .github/backend-matrix.yml
• Removed the metal: capability mappings and the concrete metal-fish-speech / metal-fish-speech-development gallery entries in backend/index.yaml
• Removed the now-unused darwin-only backend/python/fish-speech/requirements-mps.txt

Unaffected

Linux CPU / CUDA 12 / CUDA 13 / ROCm / Intel / L4T variants are untouched. Referential integrity of backend/index.yaml was verified (no dangling capability refs) and both YAML files parse.

Assisted-by: Claude:claude-opus-4-8 [Claude Code]