ieee802154/submac: avoid race condition between RX_DONE and ACK_TIMEOUT

[jia200x]

Contribution description

This PR makes adapts the documentation of ieee802154_submac_ack_timer_fired to prevent execution when ieee802154_submac_ack_timer_cancel was called. Since the SubMAC does not impose any rules on how it should be processed, this leaves the responsibility to the upper layer.

This also adapts netdev_submac_ieee802154 and tests/ieee802154_submac accordingly.

Testing procedure

It's very hard to trigger, but the consequences of this bug are more visible when using socket_zep (see #16932 (comment))

Anyway, here's the test output for tests/ieee802154_submac and examples/gnrc_lorawan.

> 2021-10-07 14:47:50,777 # Tx complete
2021-10-07 14:47:50,907 # txtsnd 00:04:25:19:18:01:BD:DB 100
> 2021-10-07 14:47:50,912 # Tx complete
txtsnd 00:04:25:19:18:01:BD:DA 100
2021-10-07 14:47:58,642 # txtsnd 00:04:25:19:18:01:BD:DA 100
> 2021-10-07 14:47:58,676 # No ACK
txtsnd 00:04:25:19:18:01:BD:DB 100
2021-10-07 14:48:00,072 # txtsnd 00:04:25:19:18:01:BD:DB 100
> 2021-10-07 14:48:00,078 # Tx complete

2021-10-07 14:59:45,788 # 1032 bytes from fe80::ac8d:fee1:20bc:404d%6: icmp_seq=999 ttl=64 rssi=-71 dBm time=140.303 ms
2021-10-07 14:59:46,645 # 
2021-10-07 14:59:46,649 # --- fe80::ac8d:fee1:20bc:404d PING statistics ---
2021-10-07 14:59:46,653 # 1000 packets transmitted, 980 packets received, 2% packet loss
2021-10-07 14:59:46,658 # round-trip min/avg/max = 132.505/147.343/169.626 ms
> ifconfig
2021-10-07 15:19:26,346 # ifconfig
2021-10-07 15:19:26,353 # Iface  6  HWaddr: 3D:DB  Channel: 26  Page: 0  NID: 0x23  PHY: O-QPSK 
2021-10-07 15:19:26,354 #           
2021-10-07 15:19:26,358 #           Long HWaddr: 00:04:25:19:18:01:BD:DB 
2021-10-07 15:19:26,365 #            TX-Power: 0dBm  State: IDLE  max. Retrans.: 3  CSMA Retries: 4 

Issues/PRs references

Discovered by @benpicco in #16932 (comment)

[jia200x]

I guess this should be backported

[benpicco]

This fixes the problem I see with socket_zep and the code makes sense.
When the ACK timeout is canceled, we should clear the flag in case it already happened.