Support catalogs

[joefreeman]

This adds initial support for a project catalog. Tasks can publish values to the catalog, read current values, and wait for the next value. Waiting works with suspense, so an execution can go to sleep until an update become available.

A catalog can be defined at the module level, it can be typed, and its path an be parameterised:

models = cf.Catalog("models/{name}")
counts = cf.Catalog[dict[str, int]]("counts/{id}")

Within a task, use .at(...) to resolve the placeholders (if needed), and .publish(...) to write a new value:

counts.at(id="words").publish({"the": 412})

If Pydantic is installed, types are validated at publish and read.

Within a task use .current() to get the latest value (relative to an execution's snapshot) - this will block (or suspend) if no value is available:

churn_rate = models.at(name="churn_rate").current()

Or use .next() to suspend execution, deferring to an execution snapshotted against the next version (when available):

models.at(name="churn_rate").next()

Catalogs are scoped to a workspace using the same inheritance rules as elsewhere.