Implement accessors to read dataset events defined as inlet

[uranusjr]

This is kind of the other side of dataset_events implemented in #38481. The inlet_events context key allows the task to access past events associated with a dataset that’s defined in the task’s inlets, like this:

@task(inlets=my_ds)
def my_task(inlet_events):
    last_event_timestamp = inlet_events[my_ds][-1].timestamp

Note that inlets is not logically related to using the dataset to schedule a DAG. An inlet dataset may or may not also be present in the DAG’s schedule. Subsequently, events accessed from inlet_events do not contain any logical filtering—all past events are simply returned with a list-like interface.

This PR implements the basic structure including a lazy list-like structure (that queries the database on-demand). I plan to add more changes in future PRs after this is merged (all targeting 2.10):

• Rename dataset_events to outlet_events for consistency.
• Allow slicing syntax e.g. inlet_events[ds][:-3].
• Some refactor to consolidate other list-like and lazy db access interfaces we provide elsewhere, most significantly LazyXComAccess.
• Add documentation on this.