fix(mhires): hysteresis on percell bg0 to damp background/letterbox $D021 flash

[kfox]

Problem

In palette_mode = "percell" (the MultiHires default), bg0 — the global %00 colour written to $D021 — was picked each frame as the raw argmax of the EMA-smoothed palette counts. On near-tied content (a mostly-black frame with a bright moment, or letterboxed/pillarboxed video whose bars quantize to black) the argmax flip-flops frame-to-frame, strobing every %00 pixel (the background and the letterbox/pillarbox bars) a different colour for a frame. It's a single instant register change — not a screen/color/bitmap write tear — so it's especially visible on a slow transport (TeensyROM serial) where the rest of the frame lags behind.

Fix

Make bg0 sticky with relative hysteresis (BG0_HYSTERESIS_MARGIN = 0.25): keep the current bg0 unless a challenger's smoothed count beats it by the margin. A sustained dominant-colour shift still moves bg0 (a real blue scene eventually turns the bars blue); a bg0 that vanishes from the frame (smoothed count → ~0) is never sticky, so it can't get stuck on an absent colour.

This is independent of palette_mode's slot allocation and only touches the percell path; cheap/vivid/grayscale already pin bg0 deterministically. This is not a perfect fix, but does reduce the bg flashes.

Scope / residual

Reduces the flash; it does not eliminate all motion artifacts. The remaining per-cell "sparkle" is the separate non-atomic screen/color/bitmap bus-DMA tear, which only host-DMA double-buffering or a lower frame rate addresses (tracked separately).

Verification

• New PercellBg0HysteresisTest (2 cases): a slight/sustained majority does not flip bg0; an overwhelming sustained change still does.
• Full suite green (1573 tests); ruff + pyright clean; pre-commit hooks pass.
• HW-verified on TeensyROM over serial — measurable improvement in the background/bar flash.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.59%. Comparing base (29487b0) to head (08d06de).
⚠️ Report is 2 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #41      +/-   ##
==========================================
+ Coverage   79.57%   79.59%   +0.02%     
==========================================
  Files          68       68              
  Lines       12861    12870       +9     
  Branches     1898     1899       +1     
==========================================
+ Hits        10234    10244      +10     
  Misses       2188     2188              
+ Partials      439      438       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.