This is because according to VMware documentation1, these are the steps to replace a failed host in a stretched cluster:

Run the compact cluster API to remove any stale data from vSAN.

Decommission the host to be removed using SDDC Manager UI or API.

Commission the replacement host to the same network pool as the removed host using SDDC Manager UI or API.

Add the newly commissioned host to the cluster using SDDC Manager UI or API.

Explanation: According to the VMware documentation, when replacing a failed host in a stretched cluster, the first step is to decommission the failed host. This should be followed by removing the host using cluster APIs, commissioning the new host with the correct network, and then adding the newly commissioned host to the cluster using cluster APIs.