feat: Python version matrix for Docker images + error-masking fix

[deanq]

Summary

• Parameterize all 4 Dockerfiles with PYTHON_VERSION build arg (default 3.11)
• Add GPU PyTorch base image mapping: 3.11 -> pytorch 2.9.1, 3.12 -> pytorch 2.10.0
• Add Makefile version matrix: build-*-versioned, build-wip-versioned, smoketest-versioned targets
• Produces 10 tagged images: GPU (3.11, 3.12) x 2 + CPU (3.10, 3.11, 3.12) x 2
• Fix error-masking in handler.py: deployed mode now raises RuntimeError instead of silently falling back to Live Serverless

Context

AE-2391: Python version mismatch between user build environment and worker runtime causes silent failures with binary packages (numpy, etc). This PR adds versioned Docker images so the SDK can select the correct runtime for the user's Python version.

Companion PR: runpod/flash (SDK-side version detection and image selection)

Test plan

• make build-all-versioned builds all 10 images locally
• make smoketest-versioned verifies correct Python version in each image
• Existing make test passes unchanged
• Handler error-masking fix verified: deployed mode raises on handler load failure

[Copilot]

Pull request overview

Adds a Python-versioned Docker image matrix (CPU/GPU, QB/LB) to prevent runtime Python mismatches with binary deps, and makes deployed endpoints fail fast when the generated handler can’t be loaded (instead of silently falling back to the Live Serverless protocol).

Changes:

• Parameterizes Docker builds via PYTHON_VERSION / PYTORCH_BASE build args and introduces versioned image tags (e.g., :py3.11-...).
• Adds Makefile targets to build/push/smoketest the multi-version image matrix.
• Updates generated-handler loading behavior to raise in deployed mode, with corresponding unit test updates.

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
Makefile	Adds Python version matrix variables and versioned build/push/smoketest targets.
Dockerfile	Switches GPU base image selection to PYTORCH_BASE arg (and adds PYTHON_VERSION arg).
Dockerfile-lb	Same as GPU QB Dockerfile, but for LB image.
Dockerfile-cpu	Parameterizes CPU base image via python:${PYTHON_VERSION}-slim.
Dockerfile-lb-cpu	Parameterizes CPU LB base image via python:${PYTHON_VERSION}-slim.
src/handler.py	Makes deployed-mode handler loading failures fatal; keeps Live Serverless path separate.
tests/unit/test_handler.py	Updates _load_generated_handler tests to expect RuntimeError instead of None.
uv.lock	Updates lockfile with multiple dependency version bumps.

---

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

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 10 changed files in this pull request and generated 3 comments.

---

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

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 10 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

tests/unit/test_handler.py:179

• test_loads_generated_handler_from_file patches handler.Path to return a tmp_path file, but _load_generated_handler() enforces handler_file.resolve().is_relative_to(Path('/app').resolve()). Since tmp_path is not under /app, this test will raise the "resolves outside /app" RuntimeError instead of returning a loaded handler. Adjust the test to bypass the /app sandbox check (e.g., mock resolve()/is_relative_to or patch the /app root used by the function) or refactor _load_generated_handler to use a patchable constant for the app root.

    def test_loads_generated_handler_from_file(self, tmp_path):
        """With valid generated handler file, loads and returns handler function."""
        handler_file = tmp_path / "handler_gpu_config.py"
        handler_file.write_text(
            "async def handler(event):\n"
            "    return {'result': event.get('input', {}).get('prompt', 'default')}\n"
        )

        with patch.dict("os.environ", {"FLASH_RESOURCE_NAME": "gpu_config"}):
            with patch("handler.Path", return_value=handler_file):
                result = _load_generated_handler()

---

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

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 10 changed files in this pull request and generated no new comments.

---

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

You can also share your feedback on Copilot code review. Take the survey.