feat: googly desk pet — face engine, mood, touch reactions, EEPROM persistence

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

DECISIONS.md

@@ -0,0 +1,25 @@
+# googly — Decisions
+
+- 2026-07-11: Big procedural face over pixel-art sprite creature — reads as alive
+  from desk distance, needs zero art assets, and partial-region redraw sidesteps
+  the 8-bit bus being too slow for full-screen animation.
+- 2026-07-11: Light mood state (b) over pure toy (a) or full Tamagotchi (c) —
+  (a) goes stale, (c)'s neglect mechanic fights weekends (always-dying pet =
+  guilt, not cute).
+- 2026-07-11: Mega 2560 as target — 256KB flash / 8KB RAM vs UNO's 32K/2K which
+  graphictest already fills to 98%. Firmware backed up before overwrite
+  (`../backups/mega-*-flash.hex`).
+- 2026-07-11: No SD dependency — SD on this shield needs soft-SPI on the Mega
+  (pins 10-13 aren't Mega hardware SPI) and the face doesn't need assets anyway.
+- 2026-07-11: Touch used only as tap/hold/pressure, not calibrated x/y — poke and
+  pet don't need position, which makes per-shield resistive calibration
+  unnecessary. Ghost-touch guard: pressure window + 2 consecutive samples.
+
+## Deferred / Rejected
+
+- Serial-dashboard generic firmware — separate idea, better fit for the
+  Leonardo/UNO (native USB on Leonardo). Not part of googly.
+- RTC / real day-night cycle — no RTC on board; sleep is inactivity-based.
+  Revisit only if a DS3231 module shows up.
+- Sound (piezo chirps) — no free pin budget to fight the shield over; also a
+  desk toy at work that beeps is a desk toy that gets unplugged.