feat: Add multi-stage multi-arch Containerfile

[FreddyFunk]

What does this PR do?

Adds a multi-stage, multi-arch Containerfile that produces two image targets: a python-export image for model quantization and ONNX export and a leaner runtime image with only the C++ binaries for engine building and inference. No GPU is required at build time, enabling CI and headless server builds. Also adds a documentation page covering build instructions, usage examples and available export tools.

Relates to #32. The file is named Containerfile rather than Dockerfile to reflect that it is an OCI-compliant container image, compatible with both Podman and Docker.

Type of change: ?
new feature, documentation

Overview: ?

Usage

# Build the lean C++ runtime image (default)
podman build -t tensorrt-edgellm .

# Build the Python export pipeline image
podman build --target python-export -t tensorrt-edgellm-export .

# Cross-build for aarch64 on an x86_64 host
podman build --platform linux/arm64 -t tensorrt-edgellm-arm64 .

# Run with GPU access (Podman CDI)
podman run --rm -it --device nvidia.com/gpu=all \
  -v ./workspace:/workspace/output tensorrt-edgellm

# Run with GPU access (Docker)
docker run --rm -it --gpus all \
  -v ./workspace:/workspace/output tensorrt-edgellm

Testing

All three image builds verified on x86_64 host with a RTX 3090 and a x84_64 host without a NVIDIA GPU :

• podman build -t tensorrt-edgellm . --> builds successfully (10.4 GB)
• podman build --target python-export -t tensorrt-edgellm-export . --> builds successfully (19 GB)
• podman build --platform linux/arm64 -t tensorrt-edgellm-arm64 . --> cross-builds successfully via qemu-user-static (10.4 GB, arm64)
• GPU passthrough validated with both Podman (CDI) and Docker (--gpus all)
• Exports and GPU accelerated inference works as expected

Before your PR is "Ready for review"

• Make sure you read and follow Contributor guidelines and your commits are signed.
• Is this change backward compatible?: Yes
• Did you write any new necessary tests?: No. Container builds are validated manually; no CI container build pipeline exists yet in this project but could be added in a separate pull request.
• Did you add or update any necessary documentation?: Yes
• Did you update Changelog?: Yes

Additional Information

I verified arm64 container builds via cross compiling but I do not have any arm64 hardware capable of connecting to a NVIDIA GPU available. I expect everything to work just fine, but I suggest that someone with access to arm64 hardware test this before merging this pull request.

[nvluxiaoz]

Thanks a lot for your MR! TensorRT Edge-LLM were committing to provide some Docker recommendations, but not with a fixed Dockerfile now as Dockerfiles need careful maintenance and infra costs. The team will re-evaluate the need for this and evaluate whether this MR should be merged.

[nvluxiaoz]

Thanks for your PR! Our team will evaluate it later on whether this MR's approach is good, or to create DockerFiles for the environment.