fix(modbus): reg 15 re-asert kazdy tick + per-charger failsafe (BUG1)

Zivy incident home-01 (TeltoCharge .16): od ~22:45 UTC 12.6. nevznikl zadny
telto journal radek (ani failed), auto jelo failsafe 8 A misto planovanych 0 A.

Root cause: reg 15 (amps) byl write-on-change proti journalu
(fn_modbus_device_state_map). Jakmile mel reg 15 radek "0 verified" a plan
dal chtel 0, NIKDY nevznikl novy prikaz -- a TeltoCharge si po vypadku
komunikace sam prepsal reg 15 na failsafe (reg 20) BEZ journal radku. Verify
cte zpet jen 'written' radky, takze tichy drift 0 -> 8 A nikdo nevidel ani
neopravil.

- reg 15 (amps to use) se zapisuje VZDY (re-asert) -- volatilni ridici
  registr, ne EEPROM; drzi verify jobu cerstvy written radek -> drift se
  zachyti a hned opravi. _split_amps_and_watchdog odděluje 15 od 19/20.
- reg 19/20 (watchdog config, EEPROM) zustavaji write-on-change.
- per-charger failsafe/timeout: asset_ev_charger.watchdog_failsafe_a /
  watchdog_comm_timeout_s (V106; default 8 A / 300 s). "Zakaz nabijeni" =
  reg 15 = 0 (protokol rev 0.5 nema samostatny enable registr).
- testy test_ev_write_on_change.py; docs teltocharge + journal + data-model.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs/03-data-model.md

@@ -175,6 +175,10 @@ CREATE TABLE asset_ev_charger (
     phases            INT DEFAULT 3,
     connector_count   INT DEFAULT 1,
     schedulable       BOOLEAN DEFAULT true,
+    -- TeltoCharge watchdog (V106): reg 19/20. Failsafe = proud po výpadku
+    -- komunikace EMS; běžný provoz řídí reg 15 z plánu, ne failsafe.
+    watchdog_failsafe_a      INT NOT NULL DEFAULT 8,    -- reg 20: 0–32 A (0 = po výpadku nenabíjet)
+    watchdog_comm_timeout_s  INT NOT NULL DEFAULT 300,  -- reg 19: s bez komunikace → failsafe
     notes             TEXT
 );
 ```