Add new Headless Dockerfile & configure GH action to build & push

[hammady]

Description

This PR introduces the following:

1. A brand new Dockerfile (docker/Dockerfile_ubuntu_headless) to build Meshroom (headless) from scratch on the latest Ubuntu and CUDA. Headless means only the meshroom python package and binaries (e.g. meshroom_batch). This is much smaller, simpler and more secure than the fully fledged Qt build. This was done because there are some use cases where one wants to embed meshroom in their Python code without requiring the heavy dependency on Qt.
2. A new GitHub action which automatically builds a docker image and pushes it to ghcr.io on every push to the specified branches. This requires a secret to be configured by the repo owner: GHCR_PUSH_TOKEN which is a PAT with the packages:write action. This can be easily changed to push to Docker Hub if desired (or both). This has been tested on my fork (check here) where I invoked the meshroom_batch command on the official monstree full dataset.
3. Removed dependency on the docker/build* scripts which is not the recommended way of invoking docker builds as they should be self-contained and independent of the host OS.

Features list

1. New Docker build for headless mode
2. New GitHub action

Implementation remarks

Here is an example Python app that mimics the meshroom_batch binary assuming only the --input and --output arguments are passed:

from meshroom.core.graph import Graph, GraphModification, executeGraph


def execute_pipeline(input_dir, output_dir, name, pipeline_file):
    # create an empty graph
    graph = Graph(name=name)

    # update attributes of the graph (input dir, output dir and pipeline to load)
    with GraphModification(graph):
        # load graph template from the pipeline file
        graph.load(pipeline_file, setupProjectFile=False)

        # get init node and initialize it with input directory
        initNode = graph.findInitNodes()[0]
        initNode.nodeDesc.initialize(initNode, input_dir, [])

        # get publish node and set output directory
        publishNode = graph.nodesOfType('Publish')[0]
        publishNode.output.value = output_dir

    # start computation
    executeGraph(graph)

execute_pipeline(input_dir='/input', output_dir='/output',
    name='someweirdnameishere', pipeline_file='/home/pipeline.mg')