fix(callbacks): wrap Oban callbacks

[solnic]

Now the Oban job exception and cron check-in telemetry handlers run through Sentry.Callback.guard/3. A crash is logged as an error with :sentry log domain and goes out as a client report with the internal_sdk_error reason, and the telemetry handler stays attached — before this, one bad callback made :telemetry detach the handler and silently ended all Oban error reporting and check-ins for the lifetime of the node.

Each callback then degrades on its own terms:

• should_report_error_callback and should_report_error_check_in_callback are treated as if they returned true
• monitor_slug_generator falls back to the slug derived from the worker name
• sentry_check_in_configuration is skipped so the check-in is sent with the options the integration inferred
• oban_tags_to_sentry_tags leaves the event with the integration's own tags (a non-map return is logged as a warning and does the same).

Logging potential crashes is also consolidated now.