expose fbrunner itself with context manager

Summary:
Right now, when we `enter` context after `get_fb_runner`, we are using the existing definition in `Runner` which causes type erasure to the super type `Runner`.
While this is fine for most cases, we lose the ability to access the `dryrun` api defined in https://www.internalfb.com/code/fbsource/[587e66c70ce546a78a71b1a22125850d1aff866d]/fbcode/torchx/runner/fb/api.py?lines=103
which has extra parameters.

For orchestration sdk to translate single job definitions as part of its dag correctly, we wish to use FBRunner's implementation of `dryrun`, so we need to be able to access `FBRunner` after entering the context manager.

Here, we add corresponding `enter` and `exit` on the FBRunner to prevent erasing the type.
All implementation still delegates to the existing super functions

Reviewed By: kiukchung

Differential Revision: D72677312

Author: bobyangyf

torchx/runner/api.py

@@ -43,6 +43,7 @@
 
 from torchx.util.types import none_throws
 from torchx.workspace.api import PkgInfo, WorkspaceBuilder, WorkspaceMixin
+from typing_extensions import Self
 
 from .config import get_config, get_configs
 
@@ -120,7 +121,7 @@ def _get_scheduler_params_from_env(self) -> Dict[str, str]:
                 scheduler_params[lower_case_key.strip("torchx_")] = value
         return scheduler_params
 
-    def __enter__(self) -> "Runner":
+    def __enter__(self) -> Self:
         return self
 
     def __exit__(