Summary

This change replaces the global "any prior Workflow\\Exception log" suppression with a replay probe.

When an exception job arrives for a missing log index, we now:

• append a synthetic in-memory exception log to the loaded workflow history
• replay the workflow in a probing mode that does not dispatch missing work or execute side effects
• persist the real exception log only if replay actually matches that exact exception on the active path

This preserves the child-exception propagation fix from #348 while avoiding the later logging regression reported in #372.

Root cause

1.0.70 changed exception persistence from "one log per index" to "only the first exception log anywhere in the workflow". That deduped stale parallel sibling failures, but it also suppressed legitimate later exceptions after:

• signal-driven retries
• consecutive caught failures at new indexes

Heuristics like "previous log is an exception" or "same replay timestamp" were still too coarse, because real workflows can catch one exception and then legitimately hit another failure without any neutral log in between.

What changed

• add a probing replay mode to WorkflowStub / Workflow
• prevent probing replays from dispatching missing activities, children, timers, side effects, or versions
• pass the failing child source class into Workflow\\Exception jobs
• only persist an exception log if the probing replay matches that exact (index, class) on the live path
• add regression coverage for:
  • signal-driven retries logging multiple sequential exceptions
  • back-to-back caught failures logging multiple sequential exceptions
  • parallel child failures still deduping to one parent exception log