The trust flag at the project level is only used to read the vector layer extent from xml, not from the provider.
This flag was not available at the vector layer and data provider level. A new QgsMapLayer reading flag to propagate the trust layer metadata for the project's read flag and a new provider flag to trust the datasource config.
Trusting the datasource config allows the provider to use estimated metadata, ensures the primary key is unique and that the detectable geometry type and SRID are the same as the requested.
This feature was developed by rldhont