BE-025 LOW: perf + style — repeated full-wtlog scans, double hunk retention, redundant re-encode, whitespace

Four perf/style cleanups from the be/ review (none change behavior):

core/discover.js:75 — row0Uri scans the ENTIRE wtlog to read row 0 (ulog.each has no early stop), and find() runs it REPEATEDLY per command → O(wtlog) per invocation for a single row.
core/loop.js:81 — _hunkSink retains every fed record in the closure recs array forever (never drained; replay could read the old HUNK log instead) → ~2× peak memory alongside the ram HUNK log.
core/emit.js:150 — flush encodes the rendered body to UTF-8 bytes then immediately DECODES them back to a string for writeStdout (a redundant round-trip per flush).
core/loop.js:144 — mintBe's wrap-defaults block is tab-indented with trailing whitespace in an otherwise 2-space file.

From the 2026-07-07 be/ review (BE/REVIEW), CONFIRMED. Method Issues.

Input

row0Uri: ulog.each needs an early-stop (row 0 is the first row); find() calls it once per command, sometimes repeatedly.
_hunkSink: pager.hunksFromLog(sink.log) shows replay can come from the log, making the JS recs copy avoidable.

row0 reads via an early-stopping first-row accessor; _hunkSink drops the redundant recs retention (replay from the log); emit writes bytes without the decode round-trip; the whitespace normalized.

Early-stop row0; drop recs double-retain; kill the re-encode; fix whitespace; suite green.