fix(native): recover graphs after relay endpoint retirement

[roodboi]

A ready native graph can outlive its foreground process after the relay removes its endpoint. Explicit same-boot recovery previously required that endpoint to remain present, leaving the retained graph inaccessible.

Add an explicitly selected absent-relay recovery path bound to the exact dead foreground process, receipt, boot, publication root and held lock. Preserve application volumes and sibling graphs; refuse changed, partial or live ownership. Retained-data removal now recognizes completed same-boot recovery while retaining explicit deletion selection and proof checks.

Validation on c76c3b6:

• Independent source review found no blocking defect; all eight hosted CI jobs passed.
• Default Rust suite: 1,095 passed / 63 ignored. All-feature suite: 1,194 passed / 92 ignored. Default and all-feature Clippy, formatting, CLI typecheck/lint and tracked-file privacy checks passed.
• Native fixture passed absent-relay recovery, changed/live ownership refusals, four interrupted-recovery boundaries, retained-data and live-sibling checks, interrupted volume removal and idempotent retries. Fixture resources were removed through managed APIs.
• Two host-filesystem tests failed during an overlapping build; their focused serial rerun and the complete default serial suite passed without changing assertions or timeouts.

Release signal: fix. Target: next. Package and real-application acceptance remain tracked by HACK-1124; this PR does not claim whole-product parity or prerelease qualification. Real-application recovery exposed an existing requirement to explicitly release verified exited or never-launched bridge reservations before live-owner recovery; automatic normalization remains a follow-up. Internal research, raw logs, runtime data and credentials are excluded.

[linear-code]

BOSS-98