feat(demo): add argus-tracker demo with composite keys, batch ingestion, and TimeUUID

[Copilot]

Scaled-down test tracker inspired by scylladb/argus demonstrating complex real-world Cassandra modeling patterns: composite/compound partition keys, TimeUUID clustering, batch ingestion, and partition-scoped queries.

Models (models.py)

• User — single PK, secondary indexes on username/email/api_token, List[str] roles
• TestRun — PK: build_id, CK: start_time DESC — "latest run for build X" pattern
• Event — compound partition (run_id, kind), CK: created_at DESC — "all errors for run X" pattern
• Notification — PK: receiver, CK: TimeUUID() DESC — chronological feed with sub-ms ordering

class Event(Document):
    run_id: Annotated[UUID, PrimaryKey(partition_key_index=0)]
    kind: Annotated[str, PrimaryKey(partition_key_index=1)]
    created_at: Annotated[datetime, ClusteringKey(order="DESC")]
    metadata: Dict[str, str]  # CQL map<text, text>

Seed pipeline (seed.py)

• Logged batches for test runs (atomic), unlogged for events/notifications (throughput)
• Partition-scoped query demo after ingestion — single-partition reads on all three key patterns
• Prepared-statement caching demonstrated implicitly via coodie's driver cache on repeated inserts
• Rich CLI output with ARGUS-PRIME panopticon theme

Scaffolding

• Makefile, pyproject.toml, smoke_test.py following existing demo conventions
• README.md with model diagrams, batch pattern explanations, query examples
• demos/README.md updated with argus-tracker entry

Plan: docs/plans/demos-extension-plan.md
Phase: 8