transformers

e737e2be - Fix pe_audio sampling rate and channel dim; canonical output keys for inkling and nemotron

Fix pe_audio sampling rate and channel dim; canonical output keys for inkling and nemotron pe_audio: PeAudioAudioProcessor defaulted to sampling_rate=16000 while the legacy extractor and every published checkpoint use 48000, so it silently resampled 48 kHz input and emitted a third of the expected samples. It also needs add_channel_dim=True, since the legacy extractor declares feature_size=1 and always returns a channel axis. inkling and nemotron_asr_streaming: both were renaming their output back to the legacy input_features / input_features_mask / attention_mask keys in _postprocess_output. They now emit the canonical audio_features / audio_features_mask like every other migrated model; consumers still reading the old names get them through the deprecated-key alias. The padded- frame zeroing each performs is unchanged. All three now match their legacy extractors bit-exactly.