Use FC16 for single-register control writes (fixes writes on SUN-12K-SG05LP3)

[iz3man]

Problem

On my SUN-12K-SG05LP3 (via the Deye WiFi/LAN logger, port 8899), all control
entities (number/select/switch) fail to write. The integration sends FC06
(Write Single Register), which the inverter silently ignores. pymodbus logs
"No response received after 3 retries", the entity briefly goes unavailable and
the value never changes. Writing the same register via the
deye_modbus.write_registers service (FC16) works immediately.
Deye's Modbus protocol document specifies FC16 (0x10) for writes.

Changes

• New option Write mode (fc16 default / fc06 legacy). write_single_register()
  sends FC16 with one value when fc16 is selected, so all existing control
  entities work without map changes.
• Writes now acquire the coordinator lock used by polling. Previously a write
  could interleave with a running poll on the shared connection (transaction_id
  mismatches, lost writes).
• Write failures are logged (register + reason) instead of being swallowed, and
  entities raise HomeAssistantError so failures are visible in the UI/traces.
• Consolidated the FC05/FC06/FC16 write helpers into one _write_call().
• write_register service stays explicitly FC06 (as documented); description updated.
• README, en/cs translations, version bump to 0.1.2.

Testing

• SUN-12K-SG05LP3: TOU SOC 1 set 15 → 16 → 15 % via number.set_value,
  confirmed via the RB readback sensor; no timeouts.
• Coordinator tested against a pymodbus TCP server that drops FC06:
  fc16 → write succeeds; fc06 → logged failure, returns False.