Add split build/test callable workflow for validation

[bob80905]

Every PR test job holds a GPU runner for the full build+test duration (7–16 min), even though 67–86% of that time is CPU-only DXC/LLVM compilation. With one self-hosted runner per GPU SKU, this means builds block GPU-dependent test work unnecessarily.

This PR adds two new, standalone workflow files — no existing workflows are modified:

build-and-test-split-callable.yaml — A split version of the callable workflow:

• Build job: Runs on any available runner of the matching architecture ([self-hosted, Windows, X64] for Intel/NVIDIA/AMD, [self-hosted, ARM64, Windows] for QC). Builds DXC + LLVM, uploads artifacts via tar.
• Test job: Runs on the GPU-specific runner (hlsl-${{ inputs.SKU }}). Downloads build artifacts, extracts with --touch to preserve ninja timestamp freshness, then runs tests.
• macOS: Unchanged — stays as a single combined job since builds are only ~1 min.

validate-split-build-test.yaml — A workflow_dispatch workflow to validate the split callable per-SKU before migrating production workflows.

Fixes #1081

Assisted by: Github Copilot