CI: migrate self-hosted runners to GitHub-hosted and add pip caching

[liuxiaopai-ai]

Fixes #89

• Replace self-hosted runner labels (linux_x64/linux_arm64) with GitHub-hosted runners (ubuntu-latest / ubuntu-24.04-arm)
• Use actions/checkout@v4 (recursive submodules) instead of cloning into /tmp
• Add pip download caching for container-based jobs by pinning PIP_CACHE_DIR under ${{ github.workspace }}
• Improve concurrency group for push+PR events and add timeouts to bound CI runtime

Note: nightly_coverage still uses the existing Aliyun registry image. If that registry is not reachable from GitHub-hosted runners, a follow-up could move the image to GHCR/quay or rebuild the environment in-workflow.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}