[d-tietjen]

The CUDA feature fails to compile because both host-registration paths call a nonexistent HostRange::len(). Register each uncovered half-open span using end - start, then resolve CUDA-only strict Clippy failures without changing the public transfer handle or copy behavior.

This focused fix targets the compact-storage branch for PR 30. Main already has the CUDA defect and additional stable-Rust diagnostics; PR 30 contains the existing stable-Rust corrections. The original standalone CI failure remains in run 37563017148.

Validation of the composed candidate f2c4ffde909172a5e504ef03f8ffc6828557257b on Linux with Rust 1.93.1 and the installed NVIDIA driver:

• Strict workspace Clippy with all targets and all features passed.
• Workspace all-feature tests completed successfully: 646 reported successes and 18 ignored tests. Eight GPU benchmark functions returned early with their performance flag disabled; they receive no measurement credit. Three actual GPU smoke tests ran with explicit enable flags.
• Python controller and provenance regressions passed: 36 shared-read, 60 storage-scenario, and 11 provenance tests.
• The original isolated runtime validation passed 27 default tests and the all-feature runtime suite with three real GPU smoke tests.
• Canonical release script and refreshed hosted CI remain in progress. No benchmark qualification is claimed.

Issue: EDEN-2424