fix(history): make Phase 4 finalization interrupt-safe (#137)

[Sootopolis]

What

HistoryApp.discoverClub's onInterrupt guard covered only Phases 2–3; Phase 4 (HistoryRun.complete + logSummary) ran outside it. ZIO's .onInterrupt deregisters once the wave block completes, so an interrupt landing in the Phase-4 window ran neither finalizer, leaving history_run stranded in its non-terminal Phase-1 state.

Fix

• Widen the interrupt guard to span Phases 2–4 as one effect.
• Gate the onInterrupt finalizer on a new finalizedRef.
• Make the success-path finalize (complete + logSummary + finalizedRef.set(true)) commit atomically (.uninterruptible).

Result — exactly one finalize on every exit, run always terminal:

• success → atomic full-stats finalize, flag flipped; a late interrupt finds the flag set and skips.
• interrupt anywhere earlier → flag false → partial-stats finalizeInterrupted.
• never both (the old code's risk was neither, not double).

Fixes

Closes #137 (surfaced during review of #138).

Testing

• Full sbt test green via pre-push (960 passed).
• testOnly history + TestJobRunner + stats + membership (143) — green, incl. TestHistoryAppNetworkOutage abort path.
• No bespoke interrupt test: deterministically hitting the Phase-4 microsecond window would be flaky (against repo's deterministic-test preference); the fix is structural and compiler-checked, existing interrupt/abort coverage unbroken.

🤖 Generated with Claude Code