fix(ci): use unique docker container names

[RerankerGuo]

PR type

• Bug Fix
• New Feature
• Document Updates
• More Models or Datasets Support

PR information

Background

The GPU CI launcher always names its Docker container swift-ci-0 or
swift-ci-1. On a self-hosted runner, another active or stale workflow
can already own that global Docker name even after the GPU lock for the
current job is available.

This caused run 30896752033
to fail before tests started:

Conflict. The container name "/swift-ci-0" is already in use

Changes

• Include GITHUB_RUN_ID, GITHUB_RUN_ATTEMPT, and the selected GPU
  slot in each CI container name.
• Use local-<pid> and attempt 0 as deterministic local fallbacks.
• Add a CPU regression test that runs the launcher with stubbed
  docker and flock commands and verifies the complete docker run
  invocation.

Verification

• PYTHONPATH=. .venv/bin/python -m unittest tests.general.test_dockerci
• PYTHONPATH=. .venv/bin/python tests/run.py --pattern test_dockerci.py
• bash -n .dev_scripts/dockerci.sh
• .venv/bin/pre-commit run --all-files
• git diff --check

Impact

GPU selection and lock files are unchanged. Only the Docker container
name changes, preventing unrelated workflow runs from colliding in the
daemon-wide name namespace.

Experiment results

Before the fix, the stubbed launcher emitted:

--name swift-ci-0

With GITHUB_RUN_ID=12345 and GITHUB_RUN_ATTEMPT=2, it now emits:

--name swift-ci-12345-2-0

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[RerankerGuo]

CI note: the failing CUDA unittest is unrelated to this PR. test_container_name_includes_workflow_identity passes in that run; the only error is ModuleNotFoundError: transformers.models.gemma3 in test_gemma3_template. Lint and the other unittest job both pass. The repository-wide resolver regression is addressed by #9862, which keeps the Transformers 4.x pin through pip install ..

[RerankerGuo]

Closing this to reduce the active review queue to the agreed maintenance capacity. The CI container-name fix is preserved, but it changes self-hosted runner infrastructure without maintainer feedback, so I am deferring it behind user-facing correctness fixes. The implementation and regression test remain at 6bca60d2449e3d3f4154e3f3e70049b5d7c6b3f5 for reopening or salvage.