MEM-042: keeper — REFSResolve leaks the ULOG map on a long URI; KEEPGetRemote double-error rarena

REFSResolve (REFS.c:545) ULOGOpen's the refs log (mmaps data, allocates idx) then, when a matched REFS row's URI components together overflow the capture arena, the refs_capture call-chain early-returns without ULOGClose — a genuine, reachable leak driven by an over-long ref URI. KEEPGetRemote leaks rarena only on a narrower double-error path. The big known keeper recursion findings (MEM-022..025) are already tracked. From the MEM sweep.

Issues

A held ULOG map / arena isn't released on a post-open error.

REFS.c:545 REFSResolve: ULOG mmap+idx held across the refs_capture chain; an arena-too-small error (long URI) returns without ULOGClose.
KEEPGetRemote: rarena leaks on a double-error path (lower severity / narrower trigger).

Blockers

None; keeper-internal. ULOGClose is null-safe — use the capture-result/close/check idiom.

Planned

Close the ULOG / free the arena on the capture-error path.

Repro first (CLAUDE §17): resolve a REFS row with an over-long URI that overflows the capture arena; assert no map leak.
Bracket the refs_capture chain so ULOGClose runs on the arena-overflow error before returning.
Free rarena on the KEEPGetRemote double-error branch.