With this commit, it's posssible to handle a request from a QgsProject without writing it to the disk.
```python
server = QgsServer()
project = QgsProject()
vlayer = QgsVectorLayer("/path/to/shapefile/file.shp", "layer_name_you_like", "ogr")
project.addMapLayer(vlayer)
query_string = 'https://www.qgis.org/?SERVICE=WMS&VERSION=1.3&REQUEST=GetCapabilities'
request = QgsBufferServerRequest(query_string, QgsServerRequest.GetMethod, {}, data)
response = QgsBufferServerResponse()
server.handleRequest(request, response, project)
```
This change ensures that searching for algorithms always returns
native algorithms before matching 3rd party algorithms
TODO: we really need to replace the toolbox tree with a proper
model and redo the sorting/filtering using a sort/filter proxy
model.
In future we can use this to optimise the preview job and shortcut
by doing lower quality/faster renders.
For now, use this flag to identify preview jobs and only apply
'updated canvas' min/max to rasters for non-preview jobs
Fixes#16988
Previously we took a harsher approach to filtering which inputs
were acceptable for child algorithm parameters. E.g. a child algorithm
with a vector layer input would only show outputs from other algorithms
which generated a vector layer output.
But this can needlessly restrict what's possible in models. E.g.
an algorithm which outputs a QgsProcessingOutputFile or
QgsProcessingOutputString could potentially be a valid source
to a vector layer. So we should allow these as possible
inputs for vector layer parameters too.
This commit adds many extra acceptable input types for
child parameters. It will probably expose "corner cases" where
algorithms may get unexpected input types, but these will
not affect existing model stability and can be fixed when
identified.
The payoff is a much more flexible modeler.
