PERF: Combine similar raster requests

[caspervdw]

This is an optimization that will lead to less raster IO in some cases, especially when raster.Smooth is used.

In some complex graphs, we have the situation that the result of raster.Smooth is compared with the original. A contrived example:

from dask_geomodeling.raster import RasterFileSource, Smooth

source = RasterFileSource("/path/to/file")
result = Smooth(source) - source

Requesting data from this result will give 2 different raster requests:

>>> result.get_compute_graph(mode="vals", width=10, height=10, bbox=(0, 0, 10, 10), projection="EPSG:28992")[0]
{'rasterfilesource_36ee9e455dda8d91541eec2dd5dbd6b0': (<function dask_geomodeling.raster.sources.RasterFileSource.process(process_kwargs)>,
  {'mode': 'vals',
   'url': 'file:///home/casper/code/lizard-nxt/var/m_37hn1_int.tif',
   'bbox': (-1.0, -1.0, 11.0, 11.0),
   'width': 12,
   'height': 12,
   'first_band': 0,
   'last_band': 0,
   'dtype': numpy.float32,
   'fillvalue': 3.4028235e+38}),
 'smooth_0655a67553d8dc3aa40d4acf491f2681': (<function dask_geomodeling.raster.spatial.Smooth.process(data, process_kwargs=None)>,
  'rasterfilesource_36ee9e455dda8d91541eec2dd5dbd6b0',
  {'smooth_mode': 'exact', 'fill': 0, 'size': [1.0, 1.0]}),
 'rasterfilesource_8de8049cd52f1027d33a5500981fd9d3': (<function dask_geomodeling.raster.sources.RasterFileSource.process(process_kwargs)>,
  {'mode': 'vals',
   'url': 'file:///home/casper/code/lizard-nxt/var/m_37hn1_int.tif',
   'bbox': (0, 0, 10, 10),
   'width': 10,
   'height': 10,
   'first_band': 0,
   'last_band': 0,
   'dtype': numpy.float32,
   'fillvalue': 3.4028235e+38}),
 'subtract_23cc6703fd2c2bf19b6563daccad808d': (<function dask_geomodeling.raster.elemwise.wrap_math_process_func.<locals>.math_process_func>,
  {'dtype': 'float32', 'fillvalue': 3.4028234663852886e+38},
  'smooth_0655a67553d8dc3aa40d4acf491f2681',
  'rasterfilesource_8de8049cd52f1027d33a5500981fd9d3')}

The two rastersource requests obviously can be derived from one single request. Can we write a generic optimization function (see https://docs.dask.org/en/latest/optimize.html) to merge similar rastersource requests?