[ni-c]

Refs #1. v1.1 fixed most of the drain; this covers what is left.

Why

v1.1 sleeps a fixed 6 hours after one failed WiFi attempt. Three problems remain:

• 6 hours is longer than the ESP8266 can sleep at once. ESP.deepSleepMax() is 2³¹ RTC ticks, about 3.5 hours.
• One router reboot at the wrong moment costs 6 hours of data.
• If WiFi works but the MQTT broker does not answer, the device stays awake 25 seconds every 5 minutes, about 4.7 %/day by the numbers in Battery drains if WiFi is not available or working #1. The extended sleep never triggers in that case.

What

• Backoff: failed attempts are counted in RTC memory. The wait doubles from 10 minutes to at most 2 hours (BACKOFF_MAX_SLEEPTIME). The first transmission after a failure returns to 5 minutes and reports connectFailures.
• Single 5-minute sleeps: a wake that only counts down goes straight back to sleep with the radio and the sensor off. No sleep ever exceeds SLEEPTIME.
• Broker timeout counts like a WiFi failure.
• Early abort: in regular-connect mode, WL_NO_SSID_AVAIL, WL_CONNECT_FAILED and WL_WRONG_PASSWORD give up at once instead of after 20 s. The quick connect keeps its 5 s fallback.
• OTA mode never waits, and a connection made in OTA mode resets the counters as well.
• Battery percentage: an ADC value below 863 used to wrap around and gave a garbage percentage. The value is now computed in integers and clamped to 0..100, so a full battery reads 100 instead of 99.
• ESPTEMP_VERSION is v1.2, and the platform is pinned to espressif8266@4.2.1 (core 3.1.2).

Tests and CI

src/main.cpp now also runs natively against small fakes of the Arduino core and its libraries (test/support/fakes). 23 Unity tests (pio test -e native) drive the real setup() and loop(): first boot, quick connect and its fallback, corrupt and unreadable RTC, the full 10/20/40/80/120/120-minute schedule, early abort per status, broker timeout, recovery, counter saturation, low-battery cutoff and its boundary, payload rounding and size, and OTA.

• Coverage is 99.4 % of lines and 98 % of branches. The one line left is the closing brace after ESP.deepSleep().
• Six deliberate mutations (no early abort, 6 h cap, no reset, no clamp, no broker backoff, OTA honouring the wait) each turn the suite red.
• The new ci.yml runs the tests with a coverage floor (90 % lines, 80 % branches) and builds the firmware: RAM 36.6 %, flash 31.3 %.
• Dependabot stays off in this repository. The actions are pinned by SHA.

Before merge

To test on hardware: switch WiFi off and watch the attempts at 10/20/40/80/120 minutes, then switch it on again (connectFailures in the next message). Stop the broker. Check normal operation and OTA.