fix(model-trainer): force LF endings for generated train script

[saisharan0103]

ModelTrainer generated sm_train.sh using Python text-mode defaults via open(..., "w"). On Windows, that translated the \n line endings in TRAIN_SCRIPT_TEMPLATE to \r\n; the uploaded script then ran inside a Linux training container, where CRLF line endings broke bash parsing before user code started.

This updates ModelTrainer._prepare_train_script to write sm_train.sh with newline="\n", so the SDK-generated script uses LF line endings regardless of the client OS.

A focused regression assertion was added to the existing generated-driver-script unit test. It reads sm_train.sh as bytes and verifies that no CRLF bytes are present while keeping the existing content assertions.

Fixes #5904

ruff check sagemaker-train/src/sagemaker/train/model_trainer.py sagemaker-train/tests/unit/train/test_model_trainer.py reports no new findings on the changed files.
Ran pytest -x locally with no new failures.