Preserve fractional replay clocks through failed-run redrive #1137

Failed-run redrive currently serializes reused event times and the initial replayed run-start anchor with whole-second toIso8601String(). Ordinary replay consumes the persisted fractional instants. A source clock ending in .123456 can therefore become .000000 after redrive and change a workflow's decision. The timer review finding is recorded on #992; Server's matching producer correction is tracked in durable-workflow/server#1215.

Preserve the stored run-start authority for the initial clock and the recorded event authority for each consumed step. Retain already serialized replayed/reused clock strings exactly on repeated redrive. Deadline and other public timestamp formatting stay outside this correction.

Acceptance requires an additional nonzero-fraction corpus fixture without rewriting the accepted whole-second fixture, an actual first/repeated redrive that fails on the predecessor, cold query/executor replay-clock equality, unchanged source history, and no newly scheduled timer work. Normal full quality, exact CI and Security review gate the source fix. Published package and SDK Worker qualification remain separate. Workflow owns this bounded follow-up on fix/992-replay-clock-precision; active Saga and advisory qualification continues independently.