feat: add 9 advanced features with performance-optimized implementations

[Sanady]

Summary

Adds 9 major new feature modules to dataforge-py, all zero-runtime-dependency, with __slots__ on every class, lazy-loaded imports, and performance-optimized hot paths. Includes comprehensive test coverage (1870 tests pass, 9 skipped for optional TUI dependency).

New Feature Modules

Phase 1: Correlated/Conditional Fields (constraints.py + data/correlations/geo.py)

• Constraint engine with dependency DAG and topological ordering
• 6 constraint types: depends_on (geographic), temporal (before/after), correlate (statistical via Cholesky), conditional (value-dependent pools), range, and custom
• Two-pass generation: independent columns batch (column-first), dependent columns row-by-row
• Dict-based field specs in Schema: {"field": "address.city", "depends_on": "country"}

Phase 2A: Time-Series Generation (timeseries.py)

• TimeSeriesSchema with trend, seasonality (sinusoidal), noise (Gaussian), anomaly injection, spiky patterns
• Interval parsing ("1h", "30m", "1d"), regime changes, missing data gaps, range clamping
• Column-first assembly with pre-computed boolean flags for maximum throughput

Phase 2B: Data Quality/Chaos Mode (chaos.py)

• ChaosTransformer with configurable injection: nulls, type mismatches, boundary values, duplicates, whitespace, encoding chaos, format inconsistency, truncation
• Integrates as post-processing step in Schema via chaos= parameter
• Pre-checks active transforms to skip no-op paths

Phase 3A: Schema Inference (inference.py)

• SchemaInferrer: analyze CSV/DataFrame/database/list-of-dicts to auto-create matching Schema
• Pipeline: type detection, semantic type detection (regex for email/phone/UUID/etc.), distribution analysis, null rate, Schema building
• Cached field aliases and pre-compiled regex for numeric detection

Phase 3B: Data Anonymization (anonymizer.py)

• Deterministic PII replacement using HMAC-SHA256 derived seeds
• Value cache for consistent mapping (same input to same output across tables)
• Format-preserving anonymization for emails, phones; streaming CSV support
• RNG swap optimization instead of forge.copy() per unique value

Phase 4A: Database Seeding (seeder.py)

• DatabaseSeeder using SQLAlchemy (optional dep) with table introspection
• Batched INSERTs with dialect-specific optimizations (PostgreSQL COPY, MySQL FK checks, SQLite pragmas)
• Auto-schema from table structure using heuristic mappings

Phase 4B: OpenAPI/JSON Schema Import (openapi.py)

• OpenAPIParser with $ref resolution, type mapping (string+format to provider field)
• Handles enum, pattern, numeric ranges, arrays, nested objects
• New regexify() method in backend.py for pattern-based generation

Phase 4C: Streaming to Message Queues (streaming.py)

• HttpEmitter (stdlib urllib, zero-dep), KafkaEmitter (optional), RabbitMQEmitter (optional)
• TokenBucketRateLimiter using time.monotonic()
• Batch emission via schema.generate(count=chunk) for throughput

Phase 5: TUI Schema Builder (tui/)

• Textual-based interactive app (optional dep) with provider browser, schema builder, live preview, export
• CLI integration via dataforge --tui

Integration Updates

• schema.py: chaos= parameter, constraint engine, stream_to/stream_to_http/stream_to_kafka methods, None sentinel optimization for standard path
• core.py: chaos=, timeseries(), infer_schema(), infer_schema_from_csv()
• backend.py: regexify() with character classes, quantifiers, alternation, escape sequences
• cli.py: --tui/--infer/--anonymize/--chaos args
• pyproject.toml: Optional extras (kafka, rabbitmq, tui, db, all)

Performance (vs main branch, same machine)

Metric	Main	Feature Branch	Change
Startup (init)	4.5 us	4.0 us	-12% faster
Startup (first field)	50.5 us	44.5 us	-12% faster
Schema generate (100K)	357,521 rows/s	376,918 rows/s	+5.4%
Schema to_csv (10K)	282,779 rows/s	333,500 rows/s	+17.9%
Schema stream_jsonl (10K)	179,469 rows/s	199,710 rows/s	+11.3%
Scalar calls	--	--	0 regressions (171/179 improved)

Test Coverage

• 9 new test files covering all feature modules
• 1870 tests pass, 9 skipped (TUI tests -- textual not installed)
• 0 failures

Files Changed

• 29 files added/modified, ~7,300 insertions
• Zero provider files modified -- all new features are additive
• All new module imports are lazy (no startup impact)