meter-closeup: consolidate nonblocking field-test and ranking review follow-ups #178

Nonblocking threads from Codex and Greptile on PR #16 at c8c666a, assessed under the 08:30Z standard. Each needs follow-up but has no verified false reported result, meaningful data loss, privacy leak, path escape, or realistic memory growth.

• 4114894510: remove the unnecessary dataset HMAC-key requirement from standalone field tests. Missing key currently causes an explicit failure before processing.
• 4114894516: add top-three tap success to field-test summary. Current counters accurately report exact OCR read status, not candidate-list success; rank is separately printed. Keep those metrics distinct.
• 4115044766: evaluate barcode-only candidates. Current code requires a digit-bearing OCR line, reducing recall; the ranking evaluation counts the miss rather than falsely declaring identification successful.
• 4115044770: evaluate stricter barcode promotion. Substring promotion can rank an unrelated identifier first, but the experiment evaluates it against labels and PORTING.md requires homeowner confirmation. No automatic correct-identity or clearance decision was established.
• 4115044778: fix FIELD_TEST.md working-directory instructions so the experiment environment is selected. Current failure is visible module discovery, not a false result.
• 4115044779: create ocr/ before a fresh standalone q45 run. The documented full sequence runs q1 first; direct q45 fails explicitly. This repeats the mkdir item in meter-closeup: follow-ups from Codex review 5328852981 #103.

Four distinct blocking defects remain in their PR threads: note scrubbing, rebuilt correction loss, stale sweep inputs, and field-test recompression. The cache-key item previously deferred in #103 is now escalated with a synthetic changed-input reproduction; it is not part of this nonblocking list.