Tune fails to set some important Trainable properties if this is igno…

…red.

I noticed it because `self.trial_id` incorrectly returns `"default"` within a tune experiment.

This PR essentially reverts a change made 2y ago: ray-project#30042

I assume it may have solved some issue back then, but now it just causes one.

Signed-off-by: Simeon Manolov <s.manolloff@gmail.com>

rllib/algorithms/algorithm_config.py

@@ -75,7 +75,6 @@
 )
 from ray.tune.logger import Logger
 from ray.tune.registry import get_trainable_cls
-from ray.tune.result import TRIAL_INFO
 from ray.tune.tune import _Config
 
 Space = gym.Space
@@ -712,11 +711,6 @@ def update_from_dict(
         for key, value in config_dict.items():
             key = self._translate_special_keys(key, warn_deprecated=False)
 
-            # Ray Tune saves additional data under this magic keyword.
-            # This should not get treated as AlgorithmConfig field.
-            if key == TRIAL_INFO:
-                continue
-
             if key in ["_enable_new_api_stack"]:
                 # We've dealt with this above.
                 continue