Canonical names should be uppercase AGENTS.*; BigQuery and Databricks deliver lowercase

#40 surfaced this: the package constant is AGENTS_SCHEMA = "agents", and
the BigQuery and Databricks writers pass it straight through, so they create
physically lowercase objects (project.agents.dbt_model) on case-preserving
engines. SPEC.md has named the schema AGENTS since the original AGENTS.sql
specification commit, Snowflake stores the objects uppercase no matter what
the constant says, and Fivetran's internal implementation of the convention
is uppercase on BigQuery as well.

The lowercase constant predates the multi-destination writers. On Snowflake,
then the only destination, unquoted lowercase folds to uppercase, so it had
no visible effect.

Plan: flip the constant to "AGENTS", and update SPEC.md section 1 to state
the canonical spelling and Snowflake's folding behavior. On BigQuery,
dataset names are case-sensitive, so existing lowercase agents datasets
stay behind while new syncs create AGENTS. That behavior change needs a
release note and probably a migration note.

Start by locating the package constant and the BigQuery and Databricks writers that pass it through. Read SPEC.md section 1, then update the canonical spelling and document Snowflake folding, the BigQuery behavior for existing lowercase datasets, and the migration implications. Done means new syncs use AGENTS and the release and migration notes cover the behavior change.