fix(api): harden healthcheck probe failure handling

[gregagi]

Summary

• add an explicit Ninja response schema for GET /api/healthcheck for both 200 and 503
• downgrade unhealthy healthcheck dependency logs from error to warning to avoid noisy Sentry issue creation from probes
• add regression tests for DB/Redis unhealthy paths to verify 503 responses and no logger.error calls

Root cause

/api/healthcheck could return (503, payload) on dependency failure, but the endpoint only declared a 200 response model. Django Ninja raised a ConfigError when serializing the 503 response, turning an expected unhealthy result into an exception.

Verification

• ENVIRONMENT=test SECRET_KEY=test SITE_URL=http://localhost DEBUG=1 uv run pytest -q apps/api/tests.py -k healthcheck
• ENVIRONMENT=test SECRET_KEY=test SITE_URL=http://localhost DEBUG=1 uv run pytest -q apps/api/tests.py