GitHub Actions: add cross-platform release and CI workflows

[alpaim]

This PR adds cross-platform release and CI workflows.

Changes Made

1. Cargo.toml

Added feature flags for different platform backends:

• cuda - NVIDIA GPU support
• metal - Apple Silicon GPU support
• accelerate - Apple Intel CPU acceleration
• mkl - Intel CPU acceleration (MKL)

Set default-features = false on candle crates to allow feature-based selection.

2. .github/workflows/release.yml

New workflow for official releases.

Trigger: Push tags matching v* (e.g., v1.0.0)

Jobs - Builds 6 platforms in parallel:

• Linux CPU (x86_64) - with MKL
• Linux CUDA (x86_64) - with CUDA 12.4
• Windows CPU (x64) - with MKL
• Windows CUDA (x64) - with CUDA 12.4
• macOS ARM64 - with Metal
• macOS x64 - with Accelerate

Release: Creates GitHub Release marked as latest: true

3. .github/workflows/ci.yml

New workflow for CI/build verification.

Trigger: Push to main branch

Jobs: Same 6 platform builds as release

Artifacts: Uploaded with commit hash in filename (e.g., vecBox-abc1234-bin-linux-cpu-x86_64.tar.gz)

4. .github/actions/get-tag-name/action.yml

Fixed versioning logic:

• Tag push → uses actual tag name (e.g., v1.0.0)
• Main branch → uses build number (e.g., v15)
• Other branches → uses branch name + build + hash

5. .github/actions/windows-setup-cuda/action.yml

• Downloads and installs CUDA Toolkit 12.4 redistributables
• Fixed directory structure preservation (now keeps bin/, lib/ structure)
• Removed duplicate libcublasLt download
• Uses PowerShell's Invoke-WebRequest (not curl alias)

6. .github/actions/linux-setup-cuda/action.yml

New action for CUDA 12.4 installation on Ubuntu.

• Installs via NVIDIA's apt repository
• Sets up CUDA_PATH and LD_LIBRARY_PATH