Implement desired approach to error handling of individual migration/rec failures

[chadlwilson]

Context / Goal

We will need to make decisions about how we want failures to be handled and recorded/reported, and which things should be retried.

Expected Outcome

Answer and implement:

• what should happen if one of the two dataset queries fail or are invalid?
• what if an individual row fails for some reason? Should the run be abandoned?
• how should transient errors be addressed? Should the code be able to retry a query in certain situations (network problems etc)
• how should the failures be reported/recorded? Details in logs? Summary error(s) stored alongside the run results with a status?
• how should we distinguish between "user" errors (e.g bad configuration of dataset query) vs "system" errors (unexpected bug in the tool/framework)
• should any (possibly partial) results be removed from the recce DB on failure?

Out of Scope

Additional context / implementation notes

[chadlwilson]

Related to #52 (dupe?)