Add NVIDIA NeMo Text Processing example

[GeoSegun]

Summary

This PR adds a new Saturn Cloud workspace template for NVIDIA NeMo Text Processing — a library for normalizing text in speech AI pipelines. No GPU or API key required; everything runs on CPU using Weighted Finite-State Transducers (WFST).

What the template demonstrates

Two core operations used in speech AI systems:

Operation	Direction	Use case
Text Normalization (TN)	Written → Spoken	Preprocess text before Text-to-Speech (TTS)
Inverse Text Normalization (ITN)	Spoken → Written	Post-process transcript from Automatic Speech Recognition (ASR)

Files added

File	Purpose
.saturn/saturn.json	Workspace recipe — CPU Large, 10Gi disk, 1-hour auto-shutoff
start.sh	Installs nemo_text_processing on first start; skips if already installed
nemo_text_processing_demo.ipynb	Pre-built notebook with 6 sections users can run immediately
README.md	Full user-facing documentation
.gitignore	Excludes logs, cache and notebook checkpoints

Notebook structure

1. Verify Installation — confirms library is ready before running anything
2. Text Normalization — numbers, dates, times, abbreviations and measurements in English
3. Multilingual TN — same operations in German and Spanish to demonstrate 15-language support
4. Inverse Text Normalization — converts ASR spoken output back to written form
5. Batch Processing — normalize_list() for processing multiple texts in parallel
6. TTS / ASR Pipeline Examples — end-to-end pre/post-processing scenarios showing real-world usage
7. Try It Yourself — open sandbox cells for users to test their own text and language

Design decisions

• No API key, no GPU — pure WFST rules baked into the library; nothing external needed
• Notebook-first — the NVIDIA documentation recommends notebooks for this library; aligns with user expectations
• Standard Saturn Python image — nemo_text_processing installs cleanly via pip on Linux x86_64
• Install in start_script — keeps the image standard; idempotent check skips reinstall on restart
• 15 languages showcased — English is primary, German and Spanish included to demonstrate multilingual capability without overwhelming the notebook

Prerequisites for users

None — just click Start.

Test plan

• Workspace deployed on Saturn Cloud — nemo_text_processing installed in ~25 seconds
• All 6 notebook sections run successfully with correct output
• TN and ITN examples produce expected results
• Multilingual normalizers (German, Spanish) work correctly
• Batch processing cell completes without error
• start.sh idempotent check verified — skips reinstall on second start
• CI validation passes (recipe schema, naming, notebook outputs cleared)
• Add entry to templates-hosted.json when promoting to the template gallery