ci: add retry for flaky network operations (cargo test, docker compose)

[TennyZhuang]

Rationale

We observed flaky CI failures caused by transient network issues:

1. cargo test failing to fetch crates from crates.io (SSL EOF, HTTP2 framing errors)
2. docker compose up failing to pull images from Docker Hub (connection timeout, registry access denied)

These failures are unrelated to code changes and waste CI time on retries.

Changes

• .github/actions/test_behavior_core/action.yaml: Wrap cargo test in a retry loop (3 attempts, exponential backoff)
• .github/services/hdfs_native/hdfs_native_cluster/action.yml: Wrap docker compose up in a retry loop (3 attempts, exponential backoff, clean up on failure)

Testing

The retry logic is conservative: 3 attempts with sleep(i * 10) backoff, which should handle most transient network blips without significantly extending CI time for persistent failures.

[tisonkun]

+1 for docker compose retey

-1 for test retry

CI can be re run with a empty commit or committer's triggering on UI.

Retry tests may hide the real issue. For certain unstable tests, we can add retry in the test cases instead.