Failed phy request not retried and may prevent DLE procedure during auto-initiation

[mtpr-ot]

When an auto-initiated phy request collides with other procedure initiated by the peer, the "Different Transaction Collision" error code is ignored, and auto-initiation continued, effectively skipping phy update.

To Reproduce
After connection is established, the central/peripheral LLCP timing must be such that the auto-initiated phy update request from peripheral is sent just before e.g. a encryption request is received by the peripheral. This will get the request in the air, and central responds with a rejection: "Different Transaction Collision".
This behavior can be observed in the simulation test basic_conn_split.sh, which is only "saved" by the central doing a phy update later.

Expected behavior
The host app expects that at least a reasonable effort has gone into executing the auto-initiated LLCPs, doing simple retries where applicable.

Impact
Link is left in 1 Mbps phy when 2 is expected.

Environment
Issue is seen on vendor implementation based on Zephyr 2.2.1, and on latest master with CI test basic_conn_split.sh.

[mtpr-ot]

Updated description to only consider "Different Transaction Collison" issue, as the "DISALLOWED" issue was fixed.

[carlescufi]

Downgrading to low because this has a very simple workaround, which is to retrigger the procedure from the application, and there are no explicit guarantees for retrial in the auto-initiated API.
I think we should create another enhancement issue covering the behavior of auto-initiated APIs when the procedure fails.