BRO-009 jab diff/commit colour: render the tok32 diff side as bg wash

jab diff:/commit: --color showed no change highlighting — only flat +/- fg lines, with word-merged text garbled — while be shows a word-level background wash. The 2-bit diff side in each tok32 (eq/in/rm) was ignored by the renderer jab used. Fix: port bro's two-pass side→bg renderer into view/bro.js (the ONE JS impl) and route both jab colour edges through it; retire the duplicate C HUNKu8sFeedColor diff branch and the hd.color() use.

Input

Context

Repro: jab commit:?<sha> --color vs be commit:?<sha> --color — jab fg +/- with merged-text garble; be a red/green word-level bg wash.
Records are identical (same text, same tok32 incl. side bits); only the COLOUR render diverged. --plain was already byte-identical (LineBased).
Four render impls existed: C bro/BRO.c bro_cell_ansi (be, correct), C HUNKu8sFeedColor diff branch (fg, used by jab via hd.color), JS view/bro.js cellSGR (pager, diff side EQ-only), JS hd.color (dump).

Goals

jab diff:/commit: --color byte-matches be --color (bro) for the diff: view (single-file + hermetic whole-tree).
ONE C diff-colour impl (bro/BRO.c), ONE JS impl (view/bro.js).

Constraints

view/bro.js is pure JS over the JABC bindings; no new binding.
Byte-exact SGR: port abc/ANSI.c ANSIu8sFeedDelta (minimal deltas) + the dog/THEME.h THEME16TBL values (NOT the JS THEME16 approximation table).
Text bytes emitted RAW (a latin1-string + utf8.Encode double-encodes multibyte — the em-dash mojibake bug).

WIP

Design decisions

Port bro_cell_ansi (side+pass→bg I/O/J/K), bro_classify_lines, bro_classify, bro_walk_hunk two-pass (rm rows then in rows), bro_row_end_pass (hide the other side's bytes) into view/bro.js as colorDiffHunk / paintDiffRowStr.
Route: renderHunkLog (dump) + pager indexRows/paintRow detect a diff hunk (hasDiffSides) and render via the JS two-pass; plain/tlv + non-diff colour stay single-sourced through the C binding.

TODOs

view/bro.js: ansi64 model + deltaSGR/resetSGR + two-pass + cellAnsi.
renderHunkLog routes diff hunks to colorDiffHunk (retire JS-2 hd.color).
pager indexRows/paintRow pass-aware (interactive tty path).
test/diff/lib/diffcase.sh --color oracle flipped graf → be (bro).
test/diff/13-color-word-wash regression (inline/split/add/del/multibyte).
C-2 → C-1 in libdog: moved bro_cell_ansi + classify/walk/row_end two-pass (+ BRO_PASS_*/bro_lineinfo/bro_emit_fn) into dog/HUNK.c+HUNK.h; HUNKu8sFeedColor now drives it (graf --color body byte-matches be); bro/BRO.c calls the libdog copies (no dup). Built jab/be/graf/bro; 472/472 C tests, 20/20 JS.

Blockers and bummers

Whole-tree diff:?<range> over the journal store still diverges (jab emits the test submodule recursion in a different order) — PRE-EXISTING diff.js issue, reproduces in --plain too; out of scope here. Own ticket TBD.

Outcome

ONE diff-colour renderer: JS in view/bro.js (jab), C in dog/HUNK.c (graf + bro + any direct-colour tool). C-2 (the fg +/- branch) and JS-2 (hd.color for diff hunks) retired.
LANDED 2026-06-26 ("BRO-009,JS-093,JS-094: colored diffs in the js pager"): be e5cf0fdd, beagle b14b895d, dog 1be9e26c, jab 842b342.