GET-010: clone of an empty/elided-refs store writes a 0-byte refs + be get escapes HOME during shard discovery

be get "file://<abs-store>?/<proj>" against a project store whose refs is empty/elided does two wrong things instead of refusing the empty source: (1) it writes a fresh 0-byte refs (and a stray nested <store>/.be/ shard) at the target, manufacturing exactly the empty-refs shape that looks like store damage — though here no live data is lost (new-empty, not truncation, so this is distinct from ULOG-001); and (2) during shard discovery it escapes HOME — even with HOME pointed at a scratch dir it resolves the shard under the real ~/.be, so a test (or a stray clone) can silently mutate the user's real store. This is the clone/secondary-wt + store-discovery path, not the in-place truncation ULOG-001 fixed. See Issues, CLAUDE, GET-004, ULOG-001.

Issues

Two defects on the empty-source clone path.

Blockers

Need a hermetic repro that can't itself touch ~/.be — chicken-and-egg with the HOME-escape; pin discovery to the test HOME first, or run under a mount namespace.

Planned

Refuse the empty source; honor HOME in discovery.