Improve RLE container workflows and command guidance

[farhann1]

Description

Follow up on the RLE Foundry integration delivered in #9334 by addressing deferred review feedback, improving local container tooling, moving publish builds to Azure Container Registry, and making CLI guidance platform-aware.

Scope

Address deferred review feedback

• Make no-argument azd ai rle show use the environment name and project endpoint from the same .azd-rle.json state file. This prevents FOUNDRY_PROJECT_ENDPOINT from redirecting a saved environment name to a different project.
  • Context: https://github.com/Azure/azure-dev/actions/workflows/event.yml#discussion_r3707676764
• Replace the remaining errors.As test pattern with errors.AsType for consistency with the rest of the extension.
  • Context: https://github.com/Azure/azure-dev/actions/workflows/event.yml#discussion_r3707676769

Correct extension usage metadata

• Update cli/azd/extensions/azure.ai.rle/extension.yaml so invoke and show advertise the optional [environment-name] argument instead of the required-looking <environment-name> argument.
• Do not edit the generated registry.dev.json entry directly. The corrected metadata should flow into the registry with the next extension release.
• Keep the skipped 0.2.0-preview registry history in mind: its registry update landed on the feature branch rather than main, and 0.3.0-preview supersedes it.
• Context:
  • [azure.ai.rle] Dev registry update for 0.3.0-preview #9426 (comment)
  • [azure.ai.rle] Dev registry update for 0.3.0-preview #9426 (review)

Use Podman for local environments

• Replace the Docker Desktop dependency used by local init/run workflows with Podman.
• Update prerequisite checks, process execution, error messages, tests, and documentation accordingly.

Use ACR remote builds during publishing

• Replace the local container build and push performed by azd ai rle publish with an Azure Container Registry remote build.
• Continue registering the resulting image with the RLE control plane after the build succeeds.
• Surface actionable authentication, registry, build, and image-registration failures.

Improve command guidance

Make init and related next-step output shell- and platform-aware. The CLI currently prints PowerShell environment-variable syntax under WSL/Linux and still refers to the old deploy command:

$ azd ai rle init echo
Created OpenEnv-style environment at: ./echo
Next steps:
  cd "./echo"
  azd ai rle run
  $env:FOUNDRY_PROJECT_ENDPOINT = "https://<account>.services.ai.azure.com/api/projects/<project>"
  $env:AZURE_CONTAINER_REGISTRY_ENDPOINT = "<registry>.azurecr.io"
  azd ai rle deploy

The guidance should:

• Use syntax appropriate for PowerShell versus Bash/WSL.
• Refer to azd ai rle publish, not deploy.
• Clearly separate the local run workflow from the cloud publish workflow.
• Avoid presenting placeholder environment-variable commands as though every user must run them immediately.

Acceptance criteria

• Deferred review comments from fix: switch to foundry backend urls #9334 and [azure.ai.rle] Dev registry update for 0.3.0-preview #9426 are addressed and covered by tests.
• extension.yaml advertises invoke [environment-name] and show [environment-name], and generated registry metadata receives the correction through the next release.
• Local environment build/run uses Podman and no longer requires Docker Desktop.
• Publishing uses an ACR remote build rather than building and pushing locally.
• init guidance is correct for PowerShell and Bash/WSL.
• Guidance uses the current publish command and clearly explains local versus cloud next steps.
• README and prerequisite documentation reflect Podman and ACR remote-build requirements.
• Relevant extension tests, linting, and cross-platform builds pass.