easyaligner ships GPU alignment, text normalization

This is the kind of infrastructure release that matters more than a flashy demo: it focuses on the pain points people hit when aligning large speech datasets in production.

• –GPU Viterbi alignment keeps long-form audio feasible in one pass, which is the real bottleneck for large preprocessing jobs
• –Reversible normalization is a strong differentiator because it preserves original formatting instead of forcing a lossy preprocessing step
• –Automatic handling of missing transcript coverage and extra leading/trailing speech makes it more practical than many “clean data only” aligners
• –Compatibility with essentially any HF Hub Wav2Vec2 CTC model broadens the usable language/model surface area
• –The companion `easytranscriber` angle is a good sign this is meant as a pipeline primitive, not a one-off toolkit