Fix Moonshine ASR tokenizer decoding

[jeevan6996]

Summary

Fix Moonshine automatic speech recognition pipelines that fail when decoding generated token IDs.

Moonshine processors provide audio feature extraction but do not necessarily include a tokenizer. The pipeline already owns the tokenizer loaded for the model, so decoding must use that pipeline-level tokenizer rather than calling batch_decode on the processor.

Changes

• Decode Moonshine outputs with this.tokenizer.batch_decode.
• Add a regression test using a processor without a tokenizer and a pipeline tokenizer.

Validation

• Direct regression check passed with a minimal Moonshine pipeline.
• Prettier check passed for the changed files.
• TypeScript build passed.
• The focused Jest command could not complete because the test harness encountered a network fetch failed during setup.

Fixes #1735