[Commerce] Add Etsy bundle recommendation capability from order-item co-purchases

[Nickel85]

Goal

Suggest bundles based on sold order item patterns, listing names, variations, and order composition.

Recommendation Examples

• Items frequently purchased together.
• Seasonal theme bundles such as patriotic, Father's Day, dog mom, or bow/cookie-cutter sets.
• Physical full-set vs individual cutter bundles.
• Digital STL bundle opportunities.

Acceptance Criteria

• Detects multi-item orders and co-purchase patterns from order-item exports.
• Groups candidate bundles by listing theme, item name terms, variations, and buyer/order behavior.
• Produces structured bundle recommendations with evidence, confidence, expected benefit, and required user review.
• Separates digital-download bundle recommendations from physical-product bundle recommendations.
• Avoids inventing products not supported by listing/order evidence.
• Logs recommendation decisions for future performance comparison.
• Includes tests for multi-item orders, single-item orders, themed item names, and variation-based bundles.

[Nickel85]

Shared memory alignment note:

#164 now defines Mongoose as the shared agent memory provider. This issue should use the Mongoose memory provider for durable decision, preference, experiment, performance, and outcome records instead of creating an isolated agent-specific memory store.

Domain-specific logic can stay here, but storage/query/audit semantics should align with #164.