[Bug] ecommerce-catalog-search agent latency exceeds 10s target (20-94s observed)

[Cataldir]

Summary

The ecommerce-catalog-search agent responds successfully but takes 20-94 seconds per request, well above the 10-second target. The 60s upstream timeout is a symptom; the root cause is compounding sequential I/O in the agent's search pipeline.

Root Cause Analysis

Four compounding latency sources identified in agents.py and ai_search.py:

#	Issue	Impact	Fix
1	Duplicate keyword search — handle() calls _search_products_keyword again after _search_products_intelligent already calls it internally as a baseline	+3-5s wasted	Return baseline as 4th tuple element; remove duplicate call
2	Sequential CRUD fetches — _resolve_ranked_products fetches each product SKU one-by-one in a for loop	+N*latency (up to 10s for 10 products)	Parallelize with asyncio.gather
3	Sequential AI Search sub-queries — multi_query_search runs each sub-query sequentially	+N*latency (up to 6s for 3 sub-queries)	Parallelize with asyncio.gather
4	Two sequential model calls — intent classification (~8s) + response generation (~14s)	~22s minimum	Deferred (architecture change)

Fixes 1-3 are code-level optimizations. Fix 4 is deferred as it requires an architectural change.

Reproduction

POST to /invoke with query Im traveling to Russia. Which clothes you have? — response succeeds but takes 20-94 seconds.

Acceptance Criteria

• handle() does NOT call _search_products_keyword a second time after _search_products_intelligent
• _resolve_ranked_products uses asyncio.gather for parallel CRUD fetches
• multi_query_search uses asyncio.gather for parallel AI Search sub-queries
• All existing tests pass (73 tests)
• No new dependencies introduced

Files Affected

• apps/ecommerce-catalog-search/src/ecommerce_catalog_search/agents.py
• apps/ecommerce-catalog-search/src/ecommerce_catalog_search/ai_search.py