raster-aggregation in matplotlib extension

#### Is your feature request related to a problem? Please describe.

I've already stated the problem on discourse [(here)](https://discourse.holoviz.org/t/how-to-use-raster-aggregation-with-matplotlib-extension/3754), but since it does not seem to get a lot of
attention there I thought I'll bring this up in a more general way on github as well.

I'm the dev of [EOmaps](https://github.com/raphaelquast/EOmaps), which already provides an integration for datashader via the matplotlib-extension.

While working on some improvements on the link between EOmaps and datashader, I've noticed
that at the moment, (at least as I understand it) the `dsshow` function of the matplotlib-extension
only allows for `bypixel` aggregations, which limits the capabilities quite a bit.

To be more precise, I've been trying to get "raster"-like aggregations (using for example "mode" reduction) working but
I could not find a way on how to do that without monkey-patching parts of the matplotlib-extension (see below)

#### Describe the solution you'd like
A clear and concise description of what you want to happen.

at the moment, the aggregation in `dsshow()` is implemented like this:
```python
canvas = Canvas(
plot_width=plot_width,
plot_height=plot_height,
x_range=x_range,
y_range=y_range,
)
binned = bypixel(self.df, canvas, self.glyph, self.aggregator)
```

however, as I understand it, a more flexible way would be to use something like

```python
canvas = Canvas(
plot_width=plot_width,
plot_height=plot_height,
x_range=x_range,
y_range=y_range,
)

# = "raster", "point", "line" etc.
# = "mean", "median", "max" etc.
agg = canvas.(, agg=)
binned = agg.compute()
```

This is just a scetch to clarify what I mean...

#### Describe alternatives you've considered

Well, I've managed to get raster-aggregation working in EOmaps by monkey-patching the `aggregate()` function of the matplotlib-extension, but I don't think that this is a proper (and sustainable) way to provide a better datashader-integration.

#### Additional context

here's what I intend to achieve (ideally without having to temper with the matpltolib-extension myself):

Start by reading the matplotlib-extension implementation of dsshow() and its aggregate() function, which the issue identifies as the current aggregation path. Compare that path with the requested raster-style aggregation and reduction options. Done means raster-like aggregations such as mode can be used through the matplotlib extension without monkey-patching it, including the EOmaps integration described in the issue.

Written by the indexing model from the issue text.