LocalPipelineSession Mutates sagemaker_client by Injecting _pipelines Attribute

# Bug: `LocalPipelineSession` Mutates `sagemaker_client` by Injecting `_pipelines` Attribute

## PySDK Version

* [x] PySDK V3 (3.x)
* [ ] PySDK V2 (2.x)

---

## Describe the bug

`LocalPipelineSession` dynamically adds a `_pipelines` attribute to the `sagemaker_client` object:

```python
if not hasattr(self.sagemaker_client, "_pipelines"):
self.sagemaker_client._pipelines = {}
```

This modifies the client instance by attaching internal state that is unrelated to the client’s intended responsibility.

Because `sagemaker_client` is typically expected to behave like a boto-style service client, mutating it in this way introduces hidden state and side effects.

---

## Why this is problematic

This approach can cause several issues:

* **Breaks encapsulation**
Pipeline state is stored on the service client rather than on the session or pipeline objects that own the state.

* **Unexpected side effects**
If multiple `LocalPipelineSession` instances share the same client object, they will also share the same `_pipelines` registry.

* **Potential attribute collision**
`_pipelines` could conflict with future attributes added to the client implementation.

* **Harder debugging and maintenance**
Attaching internal state to an external object makes the code harder to reason about and maintain.

In general, AWS SDK components avoid mutating externally provided clients and instead maintain internal state within the session or service wrapper.

---

## To reproduce

Example demonstrating shared pipeline state when using the same client object:

```python
from sagemaker.workflow.pipeline_context import LocalPipelineSession

session1 = LocalPipelineSession()
session2 = LocalPipelineSession()

client = session1.sagemaker_client

# session2 reuses the same client
session2.sagemaker_client = client

session1.sagemaker_client._pipelines["pipelineA"] = "A"

print(session2.sagemaker_client._pipelines)
```

Output:

```
{'pipelineA': 'A'}
```

Both sessions share the same `_pipelines` state because it was attached to the client object.

---

## Expected behavior

Pipeline state should be owned by the session or pipeline classes, not by the `sagemaker_client`.

A safer design would store pipelines inside the session instance:

```python
class LocalPipelineSession(LocalSession):

def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._local_pipelines = {}
```

This ensures:

* Pipeline state is isolated per session
* No mutation of external client objects
* Clear ownership of internal state

---

## System information

* **SageMaker Python SDK version**: main branch
* **Python version**: 3.9+
* **CPU or GPU**: CPU
* **Custom Docker image (Y/N)**: N

---

Comienza con LocalPipelineSession en sagemaker.workflow.pipeline_context y sigue dónde se inicializa y utiliza el registro _pipelines. Comprueba la reproducción con un cliente compartido descrita en el issue y verifica después que el estado del pipeline está aislado por sesión y que sagemaker_client ya no se modifica.

Escrito por el modelo de indexación a partir del texto del issue.