ModelTrainer (V3) does not support output_kms_key for source code uploads

## Describe the bug

In SageMaker SDK V2, the `Estimator` uses the `output_kms_key` when uploading user training scripts to S3 (see `_stage_user_code_in_s3`):
https://github.com/aws/sagemaker-python-sdk/blob/5b3b127a2d3d12a6ff0d877ddfd9ddf13527c27f/src/sagemaker/estimator.py#L1044-L1108

In SageMaker SDK V3, the new `ModelTrainer` does not apply any KMS key when uploading source code while creating the input data channel:
https://github.com/aws/sagemaker-python-sdk/blob/9101cef1a589cf2b44e73c5456dfa46c10e32691/sagemaker-train/src/sagemaker/train/model_trainer.py#L834-L953

Additionally, even when providing an S3 URI in the `SourceCode` object (instead of a local path), the `ModelTrainer` still uploads additional driver files whenever source code is specified:
https://github.com/aws/sagemaker-python-sdk/blob/9101cef1a589cf2b44e73c5456dfa46c10e32691/sagemaker-train/src/sagemaker/train/model_trainer.py#L684-L689

These uploads do not use a KMS key.

## Expected behavior

`ModelTrainer` should:
- either respect a user-provided KMS key (similar to `output_kms_key` in V2), OR
- allow configuration of a KMS key for all S3 uploads related to source code.

## Actual behavior

- Source code and driver files are uploaded to S3 without KMS encryption.
- There is no apparent way to configure a KMS key for these uploads.

## Impact

In restricted environments (like ours), S3 policies enforce server-side encryption with KMS.
As a result, `ModelTrainer` cannot be used with custom training scripts.

This blocks use cases that rely on custom code, such as MLflow serverless integration.

## Steps to reproduce

1. Create a `ModelTrainer` with a `SourceCode` object
2. Provide either:
- a local path, or
- an S3 URI
3. Observe that S3 uploads occur without KMS encryption

## Possible solution

Expose a parameter similar to `output_kms_key` in V2, or reuse existing encryption configuration mechanisms.

## Additional context

This is a regression compared to V2 `Estimator` behavior and impacts secure environments with strict S3 encryption policies.

Beginnen Sie in sagemaker-train/src/sagemaker/train/model_trainer.py bei den in der Issue genannten Upload-Pfaden für Input-Channels und Driver-Dateien und vergleichen Sie sie anschließend mit _stage_user_code_in_s3 in src/sagemaker/estimator.py. Verfolgen Sie die vorhandene Verschlüsselungskonfiguration und fügen Sie eine gezielte Testabdeckung für lokale und S3 SourceCode-Uploads hinzu. Die Aufgabe ist abgeschlossen, wenn alle quellcodebezogenen Uploads den konfigurierten KMS-Schlüssel verwenden können.

Vom Indexierungsmodell aus dem Issue-Text verfasst.