Improve interaction between storage and fetching notifications

[jbearer]

Send notifications when an object is fetched even if we fail to store it. This prevents a case where a task waiting on a Fetch blocks for longer than strictly necessary (or sometimes indefinitely). Also includes a few related improvements to fetching, including:

• retries if we fail to store a fetched object
• trigger an active fetch on decide if the payload or VID is missing (rather than waiting for the proactive scanner to discover the missing data)
• numerous additional tests

This PR does

• Moves notifications from NotifyStorage into a higher level (the Fetcher itself). See 6016137 for further discussion. Deletes NotifyStorage
• Moves UpdateAvailabilityData/update from transaction level to data source level, so we can enforce better semantics for notifications. Again see 6016137
• Makes update infallible. Instead of failing the entire update when we fail to store one leaf, we retry where appropriate, and eventually move onto the next leaf
• Adds retries with exponential backoff when storing fetched objects
• Triggers a fetch on decide when data is missing
• Adds numerous tests, including regression tests for the deadlock bug and tests for new features (backoff and fetch-on-decide)

Key places to review

• Fetching stuff: src/data_source/fetching.rs and src/fetching.rs
• Tests for new semantics: src/fetching/provider/query_service.rs