Runtime: add auto-resume-until-clear mode for transient failures

[jafreck]

Problem

• Operators currently rerun migrate manually after transient failures to progressively clear blocked tasks.

Proposal

• Add an opt-in mode that periodically resumes from checkpoint with backoff until no transiently blocked tasks remain or a stop condition is met.
• Respect retry budgets/circuit breakers and emit clear status summaries per cycle.

Acceptance Criteria

• New CLI/runtime option enables automatic resume cycles.
• Cycles stop on success, non-transient terminal failure, or max configured duration.
• Summary output includes remaining blocked/failed tasks each cycle.
• Add tests for control-flow and stop conditions.

Context reference

• Blocked task count reduced over multiple manual resume runs (5 -> 4 -> 2), indicating automation opportunity.