Auth check for HF repos + better errors

[O-J1]

Users keep foot gunning on getting access to HF repos, including more experienced users. This attempts suppress the unnecessary parts of the error chain for hf auth. Also modifies the gitignore so that we dont try to commit users presets that sit in ./training_presets (but still track presets in sub folders)

1. Launched UI
2. Selected Krea 2 preset (I dont have access to it)
3. Expectedly fails and prints out:

Traceback (most recent call last):
  File "C:\repos\OneTrainer\modules\ui\TrainUIController.py", line 247, in __training_thread_function_impl
    trainer.start()
    ~~~~~~~~~~~~~^^
  File "C:\repos\OneTrainer\modules\trainer\GenericTrainer.py", line 135, in start
    self.model = self.model_loader.load(
                 ~~~~~~~~~~~~~~~~~~~~~~^
        model_type=self.config.model_type,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<2 lines>...
        quantization=self.config.quantization,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "C:\repos\OneTrainer\modules\modelLoader\GenericLoRAModelLoader.py", line 51, in load
    base_model_loader._validate_hf_repo_access(model_names.base_model)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\repos\OneTrainer\modules\modelLoader\mixin\HFModelLoaderMixin.py", line 81, in _validate_hf_repo_access
    raise e.with_traceback(None) from None
huggingface_hub.errors.GatedRepoError: 403 Client Error. (Request ID: Root=1-6a79c730-65e25faf298604fd20a3aeda;3e346efe-45a6-4544-90a5-86fdac5ab6fb)

Cannot access gated repo for url https://huggingface.co/api/models/krea/Krea-2-Raw/auth-check.
Access to model krea/Krea-2-Raw is restricted and you are not in the authorized list. Visit https://huggingface.co/krea/Krea-2-Raw to ask for access.

AI assistance

AI-assisted. First time trying this, something small scope. CGPT 5.6 Sol. Read every line and tested, online, offline and 'I have access' to the model

[dxqb]

was this intended? this would ignore new presets I created

[O-J1]

Also modifies the gitignore so that we dont try to commit users presets that sit in ./training_presets (but still track presets in sub folders)

Intended, doesn’t do what you describe, tested, ignores only those directly in training_presets and has no recursive behaviour

[dxqb]

make it ignore just #.json then please
user configs are already saved to training_configs not _presets
training_presets could still carry a .json file directly in the folder. just not currently because I've moved all to subfolders. if a model only has 1 preset I'd probably not put it in a subfolder

[dxqb]

I think these 3 are redudant. the model loader will eventually call one the submodule functions and reach validate_hf_repo_access