Clean up fgn example for public PR

Remove cluster-specific slurm scripts (local paths), untrack FGN.md
(dev notes), add .gitignore, and fix README references.

Signed-off-by: Kashif Rasul <kashif.rasul@gmail.com>

Author: kashif

examples/weather/fgn/.gitignore

@@ -0,0 +1,9 @@
+*.mlus
+*.png
+*.pt
+*.tfevents*
+*wandb/
+rundir/
+logs/
+*.npz
+FGN.md

examples/weather/fgn/FGN.md

@@ -16,27 +16,22 @@ The current version is intentionally an MVP:
 - autoregressive inference with stochastic trajectory rollout
 - Hydra configs and a small pytest smoke test
 
-It is designed to be extended toward the global WeatherNext/FGN-style setup
-described in [FGN.md](./FGN.md).
-
-Internally, the example should follow PhysicsNeMo and Earth2Studio variable
-conventions such as `u10m`, `v10m`, `t2m`, `msl`, `z500`, `q850`, and `w300`.
-Verbose external schema names should be treated as an I/O mapping layer rather
-than the canonical training-time names.
+The example follows PhysicsNeMo and Earth2Studio variable conventions such as
+`u10m`, `v10m`, `t2m`, `msl`, `z500`, `q850`, and `w300`.
 
 ## Layout
 
 ```text
 examples/weather/fgn/
   README.md
-  FGN.md
   requirements.txt
   train.py
   inference.py
   test_training.py
   config/
   datasets/
   utils/
+  scripts/
 ```
 
 ## Quick Start
@@ -90,4 +85,4 @@ implement:
 - multi-checkpoint deep-ensemble orchestration
 - large-scale distributed training
 
-Those are documented as the next steps in [FGN.md](./FGN.md).
+Those are the planned next steps.