fix(scanner): stop the badge glyph dot from stealing the centre ellipse

[bmc08gt]

Stacked on #1303 — it needs the sweep harness that PR adds. Retarget to code/cash once #1303 lands.

The bug

A code is located by its centre badge. The badge is a disc with the Flipcash "F" knocked out of it (ic_logo_round_white, fillType="evenOdd"), and the glyph's round dot is itself a near-perfect ellipse sitting ~0.08·D from the badge centre.

Once the code is drawn large enough in frame, that dot clears the candidate filters and lands in the same 50px prune bucket as the badge. The prune then drops the wrong one:

if (dist < 50 && 2 * potential_ellipses[i].size.area() > potential_ellipses[j].size.area())

It only scans forward (j > i), so for a big/small pair it always discards the big one. The badge disappears and the search runs from the dot, which yields one finder point instead of nine.

Trace from the instrumented detector on a 1920x1080 frame:

540px code:  potential=1  candidates=1  cand0 centre=(479,269) size=85  -> 9 finder points, decoded
972px code:  potential=2  candidates=1  cand0 centre=(464,303) size=25  -> 1 finder point,  reject

size=25 at an offset down-left of centre is the glyph dot, not the badge.

This is a hard failure band, not a slowdown: hold the phone close enough and the code will not decode until you back off. It is very likely the real story behind the reports of "occasionally-slow" scanning.

The fix

Prune only true near-duplicates — the same physical circle fitted twice from the inner and outer edge of its stroke, which are comparable in area. A much smaller neighbour is a different feature nested inside, so keep both and let the search decide.

The badge sorts first and still hits on iteration 0, so the happy path is unchanged — the trace confirms the dot is never even searched.

Harness change

The sweep substituted a plain white oval for the badge, which has no glyph and therefore no dot. That is exactly why it never caught this. It now renders the real artwork.

Measured on an S25 Ultra, real ic_logo_round_white in a 1920x1080 frame:

badge	540	756	972
solid oval (old harness)	pass	pass	pass
real logo, before	pass	fail	fail
real logo, after	pass	pass	pass

Full sweep with the real badge: 18/18 decoded, stride parity intact.

iOS

The same defect is live on iOS — CodeScanner/CodeScanner/src/scanner.cpp:721 carries the identical line and CodeView fills the badge with FillStyle(eoFill: true). Being ported alongside a wider scanner-parity pass.