Add optional AlloyDB search support

[akshayjadiyanv]

Summary

This PR adds optional AlloyDB AI vector search support for the Search Agent without changing the default Milvus-backed behavior.

• Adds a new NAT function (alloydb_search) that embeds queries via NIM and runs cosine vector search against AlloyDB (pgvector + ScaNN).
• Adds a dedicated config (configs/search-alloydb.yml) for AlloyDB deployments; the default configs/search.yml remains Milvus-backed and is unchanged.
• Adds an AlloyDB seeder (scripts/seed_alloydb.py) mirroring the existing Milvus seeder.
• Registers the new function in register.py and adds required dependencies (psycopg, pgvector, google-cloud-alloydb-connector).
• Updates the agents Docker image and adds cloudbuild.yaml for native amd64 Cloud Build image builds.

Why

The blueprint currently supports Milvus for local/docker-compose workflows. This change adds an opt-in AlloyDB path for GCP deployments (for example Cloud Run + AlloyDB AI) while preserving existing Milvus defaults for adopters who do not use AlloyDB.

Usage

Default (unchanged):

nat serve --config_file configs/search.yml --port 8005

AlloyDB (opt-in):

python scripts/seed_alloydb.py
nat serve --config_file configs/search-alloydb.yml --port 8005

For Cloud Run, pass --config_file=configs/search-alloydb.yml and set AlloyDB env vars (ALLOYDB_USE_IAM_AUTH, ALLOYDB_INSTANCE_URI, etc.).

Test plan

• Confirm configs/search.yml still uses milvus_retriever + nat_retriever (default behavior unchanged)
• Run Search Agent locally with configs/search.yml against seeded Milvus
• Seed AlloyDB with scripts/seed_alloydb.py and run Search Agent with configs/search-alloydb.yml
• Verify Cloud Run deploy with configs/search-alloydb.yml and AlloyDB IAM auth (optional GCP validation)
• Confirm recommendation agent and other configs still use Milvus unchanged