Provide a stable, documented enterprise download package for on-device dictation

## Problem

[On-device dictation documentation](https://code.visualstudio.com/docs/configure/accessibility/voice#_install-the-model-from-a-local-package) says that, where network restrictions prevent automatic download, users can download the official CPU model package separately and install it with `Chat: Install Dictation Model from Local Package...`.

The documentation does not provide a URL or a supported procedure for obtaining that package. The built-in download is performed by Foundry Local through a shared Azure ML model registry. It resolves to a short-lived signed Blob URL, which is not suitable for enterprise network allowlisting or repeatable offline distribution. The NVIDIA/Hugging Face model is not compatible with the VS Code-required ONNX package.

This affects managed corporate networks where Zscaler or comparable proxies block generic Azure ML registry and dynamic Blob Storage endpoints. In those environments, downloading the package via Foundry Local elsewhere is also not an approved workaround.

## Requested change

Please provide and document a stable, first-party distribution mechanism for the VS Code-compatible CPU package for `nemotron-3.5-asr-streaming-0.6b`, suitable for enterprise deployment. At minimum, the documentation should include:

- A stable HTTPS download URL for the complete compatible package, rather than a dynamically generated SAS URL.
- Package version, supported platforms, integrity hash, and licence information.
- A documented unattended/offline deployment or import procedure for managed devices.
- The fixed hostnames and URL paths required when an enterprise must allow the download through a network proxy.

A Copilot- or VS Code-owned distribution endpoint would enable a narrow policy exception rather than an open-ended Azure ML and Blob Storage allowlist.

## Environment

- VS Code `1.137.0` (Stable)
- macOS on Apple silicon
- Default local model: `nemotron-3.5-asr-streaming-0.6b`

## Related issue

#332537 fixed the local-package file picker and was released in 1.136.0, but it did not provide the documented, enterprise-supported package download path needed to use that command.