Fix inkling feature extractor (#47349)
* fix feature extraction
* update ingling md for style and check repo.py
* modular fixes
* Fix Inkling modeling-structure violations (TRF016, TRF019)
- Remove dead do_resize flag from InklingImageProcessor: the processor always
resizes (patch division) and forbids do_resize=False, so the flag had no effect
- Empty InklingProcessorKwargs._defaults; the load_audio_backend default moves
to processor_config.json on the hub per TRF019
* Add Inkling configs to config-docstring checkpoint ignore list
Inkling's 4 configs are dataclass-style (@strict) without per-field docstrings;
follow the GraniteConfig/Qwen3MoeConfig precedent of skipping the checkpoint
docstring requirement rather than adding @auto_docstring (which would require
documenting ~24 model-specific fields).
* Allow Inkling config-only attributes in check_config_attributes
InklingTextConfig's local_layer_ids/mtp_local_layer_ids drive layer_types in
__post_init__, and the MTP/MoE fields are carried for checkpoint round-trip;
none are referenced by the modeling forward, so allow them explicitly.
