Native support for non-text column types (JSONB, ARRAY, BYTEA, UUID)

[cl8dep]

Summary

All sanitization strategies operate by casting column values to ::text. Columns of type JSONB, BYTEA, ARRAY, UUID, range types, and others are not explicitly handled and may produce incorrect results or runtime errors when cast.

Current Limitation

For example, hash strategy generates:

encode(sha256("column"::text::bytea), 'hex')

On a JSONB column this produces a hash of the JSON text representation, which is semantically correct. But BYTEA columns would produce a hex-encoded hash of the binary's text encoding, not the binary itself.

faker strategies assign text values that may fail type casting on strictly-typed columns.

Proposed Approach

1. Type-aware SQL generation in each strategy: detect column data type during validation and emit type-appropriate SQL (e.g., sha256(column) directly for bytea, uuid_generate_v4() for UUID, '{}'::jsonb for JSONB null replacement).
2. New faker methods for typed columns: faker: uuid, faker: jsonb_empty, faker: bytea_empty.
3. Validation warning when a strategy is applied to a column type it does not natively support.

Affected Types

Type	Current Behavior	Needed
JSONB	Casts to text, works but lossy	Native JSON manipulation support
BYTEA	May corrupt binary data	Direct binary handling
UUID	Works via text cast	Native uuid_generate_v4()
ARRAY	Likely fails or corrupts	Per-element sanitization
Range types	Unsupported	New strategy or raw_sql

Acceptance Criteria

• validate reports a warning when a strategy is applied to an unsupported column type
• hash strategy works correctly on BYTEA columns
• null_value works correctly on all nullable types
• faker: uuid generates valid UUIDs for UUID columns
• faker strategies that return text emit a validation warning on non-text columns