E4-5a (band 3): fram, settings and rtc units #270

Epic: #212 · Parent: #264 — band 3, split per domain as its body anticipated ("expect this to be split into one child task per domain once band 2 has landed and the harness is proven"). Created under the standing grant the owner gave on 2026-09-07; the composition is for review.

First of five band-3 children. Three units, each taking its state with it (no globals.h, Q6 ruling), in this order because each includes the previous:

settings persists through FRAM, so it includes fram.h; syncRTCFromSystemTime calls mktimeUTC, so rtc includes settings.h. Twelve of the 34 remaining cut-set variables leave src.ino here (the settings nine, plus fram, framHeader, framAvailable).

Not moved: framFlushToSD flushes the battery and weather buffers, so it consumes two domains this task does not publish; it stays until the weather child lands.

Why not webServer first, as the text of #264 says. The relocation graph after band 2 shows webServer referenced only by the 23 web handlers, initWebServer and loop, with initWiFi calling initWebServer. A web unit moved before the domains its handlers read would need those handlers declared in a header while they stay in src.ino — the reach-back pattern the ruling forbids — or would own one variable and no code. The cut set says which variables must be owned; the graph says the order they can be. Web goes last, when everything it reads is published, and the acceptance of #264 ("every one of the 35 defined in exactly one unit") is met at that point.

Acceptance

• scripts/verify_extraction.py PASS for the task: no unexpected symbol, every size delta attributed, section deltas explained
• pio run -e feather_s3 and arduino-cli compile --fqbn esp32:esp32:adafruit_feather_esp32s3 src/ both succeed; pytest scripts/ passes
• nm on src.ino.cpp.o no longer defines any variable the three units own
• Record appended to docs/god-object-inventory.md