#10629: feat(sherpa-onnx-tts): enables sherpa support for rockchip-based CPUs using and .rknn models

# Summary Sherpa onnx has native support for running on Rockchip based NPUs when configured. (docs [here](https://k2-fsa.github.io/sherpa/onnx/rknn/index.html?highlight=rockchip)) Official pre-trained models for Rockchip NPUs can be loaded as long as the `--provider` arg is set to `rknn` and the skill logic can load `.rknn` file extensions (docs [here](https://k2-fsa.github.io/sherpa/onnx/rknn/models.html#sherpa-onnx-rk3588-streaming-zipformer-small-bilingual-zh-en-2023-02-16)). This PR enables this skill to run such models. ## Changes Scans directories for `.rknn` model extensions. Pass `--provider=rknn` param to cli when an `.rknn` model is found. <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> - Extend `sherpa-onnx-tts` model auto-detection to include `.rknn` files alongside `.onnx`. - When an `.rknn` model is selected, attempt to pass a `--provider` argument to `sherpa-onnx-offline-tts`. - Refactor CLI invocation to build an argument array (`cmd`) before calling `spawnSync`. - Net effect is intended to enable Sherpa ONNX TTS on Rockchip/RKNN runtimes, but the provider argument construction currently prevents the flag from being applied correctly. <h3>Confidence Score: 3/5</h3> - This PR is close to safe to merge, but currently fails to pass the intended `--provider` flag so RKNN support won’t actually work as described. - Change scope is small and localized to one wrapper script, but there is a definite runtime logic bug in argument construction that breaks the feature’s primary purpose. Once the provider argument is fixed, risk should be low. - skills/sherpa-onnx-tts/bin/sherpa-onnx-tts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->