V151: supplier-info source/primary columns + CRM email normalization.

Completes the V149 phoenix_kit_cat_item_supplier_info junction with the two columns the merged phoenix_kit_catalogue sourcing layer (catalogue PR #44) reads and writes — without them every junction INSERT/UPDATE crashes on an undefined column:

• supplier_source — 'crm_company' | 'crm_contact' | 'local' (CHECK). Disambiguates the polymorphic soft supplier_uuid (a CRM company, a CRM contact, or a local cat_suppliers row) so the resolver and the phoenix_kit_catalogue.audit_supplier_refs consistency report route lookups without a trial cascade.

• is_primary + partial-unique index (at most one primary row per item). The catalogue context auto-promotes the first linked supplier and offers "make primary"; Suppliers.primary_for_item/1 backs the warehouse resolver.

The V146 scalar phoenix_kit_cat_items.primary_supplier_uuid is left untouched (V149's design keeps it); the merged catalogue schema simply no longer maps it.

Also normalises the CRM party email columns to citext (phoenix_kit_crm_contacts.email, phoenix_kit_crm_companies.email): case-insensitive matching is a prerequisite for the CRM v2 backfill (match-by-email) and the user↔contact bridge (connect_user/2 finds or creates auth users by email). citext is a core dependency since V01 (phoenix_kit_users.email), so ensure_extension!/1 is a no-op on any existing install.

citext conversion cost: ALTER COLUMN … TYPE citext takes an ACCESS EXCLUSIVE lock and rewrites the table — plan a maintenance window for large CRM tables. Neither email column carries an index or a unique constraint anywhere in the chain, so no index rebuild happens and no latent case-duplicate can surface as a violation; if a unique index on email is ever added later, case-variant duplicates must be resolved first (citext compares case-insensitively).

All operations are idempotent.