Install RLE SDK wheels from public registry in sample containers

[farhann1]

Install azure-ai-projects[rle] in the four sample containers using public SDK wheels from Azure/azure-sdk-for-python#49488, without SDK source-fork access.

• Each sample includes its own local installer; tests keep the four copies identical. There is no separate repository-root installer or external installer download.
• Default: ENV AZURE_AI_PROJECTS_VERSION=latest. Edit that line to pin an available SDK version. Missing versions and checksum mismatches fail explicitly.
• Git-source ADD refreshes the public registry and wheels during normal rebuilds. The installer verifies SHA-256 and installs the selected local wheel with pip, without downloading it again.
• Local server development can use the same installer with the remote registry.
• Existing dependency settings, sample-local build contexts, server behavior, and manifests are preserved.

Available versions | Setup documentation

Review and validation

• Reviewed the complete PR against its base. Removed the obsolete fifth installer, test-only download option, historical dev-release test cases, and repetitive documentation. Restored the original Harness requirements COPY forms.
• 49 targeted tests passed, covering the four copies, Docker wiring, version selection, local-wheel reuse, checksum failures, missing wheels, and remote installation.
• All four current sample installers freshly installed the real 2.7.0b1 SDK with networking disabled in previously built sample images with dependencies present. Each real app import succeeded.
• The latest BYOH sample completed a full Docker build, installed SDK 2.7.0b1 from the locally fetched wheel, and served /health with HTTP 200. The build used the existing Microsoft package-index override because this environment blocks public PyPI TLS.
• Two unrelated, pre-existing manifest-schema assertions were excluded. No server, dataset, or manifest migration changes are included.

The SDK maintainer-tooling PR is separate and unchanged.