Fix vNext memory commit schema aliases

[samrusani]

Summary

• Normalize common agent-supplied quote memory labels to schema-backed vNext values before Postgres writes.
• Publish canonical memory_type/domain/sensitivity enums through the MCP tool schema and convert Postgres check violations into controlled MCP tool errors.
• Update agent docs so quote saves use memory_type=semantic and domain=learning instead of invented labels.

Upgrade Overview

Protected Areas

• continuity APIs

Compatibility Impact

MCP callers that send canonical vNext memory labels continue to behave the same. Common quote aliases such as memory_type=quote and domain=quotes now normalize to schema-backed values instead of surfacing a Postgres CheckViolation.

Migration / Rollout

No database migration is required. Roll out by deploying the API/MCP code and restarting any long-running alice-core MCP process so agents load the updated schema and validation path.

Operator Action

If an agent previously hit repeated alice_vnext_commit_memory CheckViolation errors, restart the agent or MCP session after deploy and retry with canonical values. For quote saves, use memory_type=semantic and domain=learning.

Validation

Validated locally with the full Python unit suite, the agentic-memory-commit smoke, API health, and whitespace checks.

Rollback

Revert this PR if the MCP enum normalization causes unexpected caller behavior. Existing database rows are unchanged because this patch does not alter persisted schema or run migrations.

Validation

• ./.venv/bin/pytest tests/unit -q
• ./.venv/bin/alicebot vnext smoke agentic-memory-commit
• curl -sS http://127.0.0.1:8000/healthz
• git diff --cached --check

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.