warn that speaker id is unsupported for batch cpu mode

[dln22]

This PR adds a warning that the SpeakerID feature is unsupported for Batch CPU mode.

[vercel]

Someone is attempting to deploy a commit to the Speechmatics Team on Vercel.

A member of the Team first needs to authorize it.

[petemomo]

Reviewed against the docs style guide, terminology, and the product-architecture context, and verified all four changes in a local build.

Solid, focused change — the limitation is stated consistently in substance across all four pages, cross-links are correct, every internal link resolves, and the warning body plus the deployments-table footnote both render correctly.

Main thing to know: the warning's custom title doesn't render (details inline) — but that's a pre-existing gap in the swizzled Admonition component, not something this PR introduced, and the body stands alone. The rest are small consistency/style items with suggestions inline. Nothing here blocks merge.

On the core claim (speaker identification unsupported on the CPU batch container, requires GPU, Realtime OK on both): I couldn't verify it against the project reference files (product-architecture.md doesn't cover the CPU/GPU support matrix), but it doesn't conflict with them — accepting on author/SME authority.

[giorgosHadji]

LGTM from tech side