Optimize feature requests when identifying

When identifying on the map canvas, be it single tap or tap and hold, a feature request is performed which fetches all attributes of all nearby features .
Most of the times this does not matter, but if a layer has an expensive attribute (eg virtual field with relation_aggregate expression) then the issue is apparent.

This can be optimized by fetching only what is needed, and only fetch the whole feature when we need to display its form:

• single tap: FID and geometry are only needed for the proximity check
• tap and hold: fields used in the display expression are only needed (no geometry)

Trace map-canvas identification for single taps and tap-and-hold, then inspect where feature requests are constructed. The change is done when single taps fetch only FID and geometry, tap-and-hold fetches only fields needed for the display expression, and the full feature is fetched only when its form must be displayed.