[Commerce] Add Etsy ad candidate recommendation from sales and listing evidence

[Nickel85]

Goal

Suggest which listings are candidates for Etsy ads using imported sales and listing evidence before full Etsy Ads performance import exists.

Signals

• Recent sales velocity.
• Listing performance score.
• Bundle/deal recommendation status.
• Seasonal relevance.
• Physical vs digital listing type.
• Average order value and net contribution proxy.
• Variation demand.

Acceptance Criteria

• Produces ad candidate recommendations with rationale, evidence, confidence, and suggested test duration.
• Distinguishes scale-up candidates from test candidates and avoid-ad candidates.
• Explains why a listing should not be advertised when economics, discounts, or weak sales evidence do not support it.
• Integrates with existing ad decision log and experiment plan issues rather than bypassing them.
• Does not claim ROI without ad spend/click/conversion data.
• Includes tests for strong sellers, low-evidence listings, coupon-heavy listings, seasonal listings, and digital vs physical listings.

[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.