feat: add SMS short-text de-identification

[maziyarpanahi]

Description

Adds an SMS-scale short_text de-identification preset and export adapters for RapidPro-shaped JSON and generic CSV helpdesk logs. The path preserves message offsets and source schemas, pseudonymizes contact values, coarsens timestamps, and processes large record streams in bounded batches.

Type of Change

• New feature (non-breaking change which adds functionality)
• Documentation update
• Performance improvement
• Test addition/improvement

Changes Made

• Add code-mixed, no-punctuation-tolerant rules for African phone numbers, shortcodes, IDs, honorific-led names, and SMS abbreviations.
• Add JSON, CSV, and iterable record handlers with deterministic per-run contact hashing, timestamp coarsening, and batched de-identification.
• Add a 600-message synthetic recall and no-leak benchmark, export round-trip and offset-fidelity tests, fixtures, an example, and user documentation.
• Preserve leading and trailing source whitespace through the pipeline and batch paths without changing the default clinical-note behavior.

Testing

• The 600-message corpus reaches complete seeded phone/name/ID redaction and higher recall than the default preset.
• RapidPro JSON and generic CSV round trips preserve schemas, row counts, flow fields, and consistent contact pseudonyms.
• The focused SMS suite passes: 6 passed.
• The full repository suite passes on current master: 6,045 passed, 48 skipped.

Additional validation:

• make format
• make lint
• make format-check
• make type-check
• pre-commit run --files <changed files>
• make docs-build
• python3 -m build

Documentation

• Added the SMS short-text guide and examples navigation.
• Added docstrings to new public functions and classes.
• CHANGELOG.md is unchanged; release notes can include this feature when it ships.

Code Quality

• Canonical formatting, lint, type, pre-commit, and strict documentation checks pass.
• The complete diff was reviewed.
• Error and summary paths avoid raw contact values and message text.
• No new warnings are introduced.

Dependencies

• No new dependencies were added.

Checklist

• Read the contributing guidelines.
• Used clear, focused commits.
• Kept the branch scoped to OM-852.

Related Issues

Closes #1429.

Screenshots/Examples

See examples/sms_deid_helpdesk_logs.py and docs/sms-short-text-deid.md for runnable JSON and CSV usage.

[maziyarpanahi]

Ready to merge. I rebased both commits onto current master and verified the complete OM-852 acceptance surface: the 600-message synthetic corpus has complete seeded phone/name/ID redaction and higher recall than the default preset; RapidPro JSON and generic CSV round trips preserve schema and flow fields; contact pseudonyms are stable within a run; batching stays bounded; and whitespace/offset fidelity is preserved without changing the default clinical-note path.

Validation is complete: the focused SMS suite passes, the full local suite passes (6,045 passed, 48 skipped), formatting/lint/type checks, all changed-file pre-commit hooks, strict docs, and package build pass. Hosted CI is fully green across Linux, macOS, Windows, security/SBOM/policy checks, and amd64/arm64 container smoke tests; the one macOS timing flake cleared on rerun without code changes. The diff and PR metadata are clean of prohibited vendor/assistant names and co-author trailers. This closes #1429.