Create developer testing environment manifests with conda-forge support

[anth-volk]

We need a canonical developer/testing environment path for cases where the normal uv/PyPI lock is not enough, especially Intel macOS where modern PyPI torch wheels are unavailable but conda-forge can provide compatible pytorch builds.

This should replace the current ad hoc no-torch fallback work from #146, #171, and #175. Those failures were symptoms of not having a supported environment route for PE-US rebuild smoke/runtime work on Intel macOS, not a reason to make the normal rebuild path pretend torch is optional when microcalibrate and related packages need real tensor support.

Scope:

• Add a dedicated developer testing environment folder, for example envs/ or developer-envs/.
• Include environment manifests for package sources beyond uv/PyPI, including conda-forge.
• Add an Intel macOS setup route that installs Python and pytorch from conda-forge, then installs this repo and its extras with pip inside that environment.
• Document when developers should use the normal uv path versus the conda-forge-backed path.
• Include validation commands for the PE-US rebuild CLI/help path and at least one smoke runtime path.
• Avoid maintaining a no-real-torch Intel Mac runtime as a supported path unless a specific command truly does not need torch.

Acceptance criteria:

• A developer can create the documented conda-forge-backed environment from a checked-in manifest.
• The documented setup supports the repo Python version range and installs a real pytorch package on Intel macOS.
• The docs make clear that uv remains the normal path where PyPI wheels are available.
• The PE-US rebuild smoke instructions no longer rely on stubbing or omitting torch for runtime work.