feat: add source-aligned text normalization APIs and CLI

[HungerDeng]

Summary

Add source-aligned text normalization for compiled NeMo FST grammars and expose it across the native and Swift interfaces. Callers now receive the complete normalized sentence together with source ranges, original text, normalized span text, and semantic classes, enabling downstream processing without string-diff heuristics.

Changes

• Preserve input progress through the selected classifier FST path and return aligned spans with stable UTF-8 byte offsets.
• Add Rust APIs for English, French, Spanish, German, Chinese, Hindi, and Japanese.
• Expose the alignment through the C ABI, Swift wrapper, and WASM binding.
• Add a SwiftPM nemo-tn-aligned JSON/JSONL CLI.
• Select the host Rust archive in SwiftPM so the existing and aligned CLIs link statically on Apple Silicon and Intel Macs.
• Document the API, build flow, and CLI output.

Testing

• cargo test --features fst-engine
• cargo check --locked --features ffi,fst-engine
• cargo fmt --all --check
• cargo clippy --all-targets --features ffi,fst-engine
• swift build and swift build --product nemo-tn-aligned
• CLI smoke tests for sentence arguments and JSONL stdin
• git diff --check

Notes for Reviewers

The per-span normalized value is direct verbalizer output; sentence-level spacing and punctuation cleanup are represented by the top-level normalized field. Offsets are half-open UTF-8 byte ranges into the original input.