Improve error formatting for deeply nested errors with large inputs

[jimsynz]

Related to ash-project/reactor#280

When Reactor returns errors, the default output can be overwhelming because:

1. Nested errors repeat large input data at multiple levels
2. The #Splode.Stacktrace<> is opaque in inspect output
3. Users don't know how to extract a readable stacktrace

Example from the reactor issue - an error struct that's hundreds of lines because inputs are repeated at each nesting level.

Ideas wanted:

• Should Inspect implementation truncate/summarise large inputs?
• Should there be a helper to format just the stacktrace chain?
• Better docs on extracting stacktraces from Splode errors?

[zachdaniel]

Commented on the original issue, but I think the concept is that Exception.message(error) should yield a decent result?