Update link to image in Fabric PPO example (Lightning-AI#17360)

Author: awaelchli

examples/fabric/reinforcement_learning/README.md

@@ -91,7 +91,7 @@ In this example we have gone even further leveraging the flexibility offered by
 The architecture is depicted in the following figure:
 
 <p align="center">
-  <img src="./imgs/fabric_decoupled.png">
+  <img src="https://pl-public-data.s3.amazonaws.com/assets_lightning/examples/fabric/reinforcement-learning/ppo_fabric_decoupled.png">
 </p>
 
 where, differently from the previous example, we have completely decoupled the environment from the agents: the *rank-0* process will be regarded as the *Player*, which runs `M+1` independent copies of the environment (*Env-0*, ..., *Env-M*); the *rank-1*, ..., *rank-N* are the *Trainers*, which contain the agent to be optimized. Player and Trainer share data through [collectives](https://lightning.ai/docs/fabric/stable/api/generated/lightning.fabric.plugins.collectives.TorchCollective.html#lightning.fabric.plugins.collectives.TorchCollective) and thanks to Fabric's flexibility we can run Player and Trainers on different devices.