Clean up Dockerfiles and add tag-triggered image publishing

[foolnotion]

Summary

• Strip the accumulated dead comments from baseDockerfile (old FROM alternatives, Nvidia GPU block, commented proxy config); combine sed + apt into a single RUN layer. No functional change to the image.
• Add ARG BASE_IMAGE=srbench/base to alg-Dockerfile and algorithms/tir/Dockerfile so that CI workflows can inject the published registry tag via --build-arg BASE_IMAGE=... without breaking local builds (the default remains srbench/base).
• Add .github/workflows/docker.yml to publish the base image and all per-algorithm images to DockerHub when a version tag is pushed (v*). Publishing is intentionally not triggered on every push to master — only on explicit releases.

Test plan

• baseDockerfile builds successfully (docker build -f baseDockerfile -t srbench/base .)
• alg-Dockerfile builds with --build-arg BASE_IMAGE=srbench/base for both gplearn (pure pip) and operon (native extension)
• bash test.sh passes all tests inside both containers (test_algorithm.py 3/3, test_evaluate_model.py 1/1) for gplearn and operon

[Copilot]

Pull request overview

This PR streamlines the project’s Docker build setup by cleaning up the base image Dockerfile, making algorithm builds configurable via a BASE_IMAGE build arg, and introducing a tag-triggered GitHub Actions workflow intended to publish Docker images on release tags (v*).

Changes:

• Simplify baseDockerfile by removing large commented blocks and consolidating sed + apt into a single RUN layer.
• Add ARG BASE_IMAGE=srbench/base to algorithm Dockerfiles so CI can inject a published base image reference.
• Add .github/workflows/docker.yml to build and push base + per-algorithm images when a version tag is pushed.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
baseDockerfile	Removes dead comments and consolidates apt setup into one layer.
alg-Dockerfile	Adds configurable BASE_IMAGE for CI-driven base image selection.
algorithms/tir/Dockerfile	Adds configurable BASE_IMAGE for the tir custom Dockerfile.
.github/workflows/docker.yml	Introduces tag-triggered DockerHub publishing workflow for base + algorithm images.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[foolnotion]

The 9 CI failures are all pre-existing issues unrelated to this PR. They were triggered because changing baseDockerfile and alg-Dockerfile correctly sets rebuild-all=true in ci-docker.yml, causing all algorithm images to be rebuilt and exposing failures that already exist on master:

Algorithm	Failure	Root cause
brush	Build fails	CMake/C++ compile error in pybrush wheel
bsr	test_evaluate_model	Sympy TypeError: Pow ^ Integer
e2et	test_evaluate_model	NameError: name 'model' is not defined in regressor
eql	test_evaluate_model	ModuleNotFoundError: No module named 'eql'
feat	test_evaluate_model	ImportError: libshogun.so.18 missing
nesymres	test_evaluate_model	Missing pretrained checkpoint nesymres_100M.ckpt
pysr	test_evaluate_model	AttributeError: _validate_data (sklearn API change)
sklearn	test_evaluate_model	ModuleNotFoundError: No module named 'methods.sklearn'
tpsr	test_evaluate_model	Missing pretrained model model1.pt

These should be tracked and fixed separately — happy to open issues for each if that's useful.

Also addressed the review feedback: image naming now matches make_docker_compose_file.sh convention (ORG/base, ORG/<alg>), versioned tags are published alongside :latest, and permissions: contents: read is set.