Share embedded function execution script template between Trainer and Spark

## Summary

Trainer and Spark currently maintain separate templates for embedding Python function source code into the container at runtime.

The Spark implementation uses a quoted heredoc delimiter to avoid shell expansion and command substitution, while Trainer uses a similar template with a plain `EOM` delimiter. The overall execution flow is conceptually the same, but the implementations have diverged.

## Proposal

Move the embedded function execution script template into a common module so both Trainer and Spark reuse the same implementation.

The shared helper should:
- Generate the embedded Python execution script.
- Be reusable by both Trainer and Spark.
- Preserve the robustness of the quoted heredoc delimiter (or another agreed-upon implementation).
- Reduce duplicated logic across the two clients.

## Benefits

- Consistent behavior between Trainer and Spark.
- Less duplicated code.
- Easier maintenance for future improvements.

ref- https://github.com/kubeflow/sdk/pull/629#discussion_r3634877066