nrf52840: unexpected PHY behavior?

While doing a sequence of energy measurements, I noticed some strange behavior on the `nrf52840dk` (`nrf52840`). While running a simple application that sends out legacy advertisements every 500ms, I can see CPU activity between two advertisement events where there should be none. This happens in non-deterministic intervals.

Trace A (ADV interval: 500ms, payload 3 byte, non-connectable, non-discoverable):

Trace B (ADV interval: 500ms, payload 3 byte, undirected-connectable, general discoverable)

Running the exact same application on the `nrf52dk` (`nrf52832`) does not show this behavior, on that board the trace looks as expected. So from that I think this might be a PHY related issue on the `nrf52840`?!

I also wonder if this could be related to #767?

Background infos:
- running NimBLE (dba097972d28701c10b9af2a48af7d924c9cfca3) @ RIOT (35b6ccedf31f10a5f8e4f97609ad5b10c28bdc34)
- the application simply calls `ble_gap_adv_start()` with the following params:
```
struct ble_gap_adv_params adv_params = {
.conn_mode = BLE_GAP_CONN_MODE_NON, // or BLE_GAP_CONN_MODE_UND for trace B
.disc_mode = BLE_GAP_DISC_MODE_NON, // or BLE_GAP_DISC_MODE_GEN for trace B
.itvl_min = BLE_GAP_ADV_ITVL_MS(500),
.itvl_max = BLE_GAP_ADV_ITVL_MS(500),
};
```
- No additional (user) threads
- NimBLE is configured as stated in `porting/targets/riot/syscfg.yml` with the following changes:
```
CFLAGS += -DMYNEWT_VAL_BLE_LL_TX_PWR_DBM=0
CFLAGS += -DMYNEWT_VAL_BLE_LL_MAX_PKT_SIZE=251
CFLAGS += -DMYNEWT_VAL_BLE_LL_CFG_FEAT_DATA_LEN_EXT=1
CFLAGS += -DMYNEWT_VAL_BLE_L2CAP_COC_MPS=200
```

Reproduce the two 500 ms advertising configurations on the nrf52840dk and compare them with the nrf52dk using the application and settings in the report. Start at ble_gap_adv_start() and porting/targets/riot/syscfg.yml, then compare the behavior with issue #767; done means identifying the source of the extra CPU activity or documenting whether it is PHY-specific.

Written by the indexing model from the issue text.