Add user guide for working with large time-series datasets

[droumis]

supersedes #1205

This adds a notebook that explains the different ways of working with large time-series datasets with holoviz

[droumis]

Not sure why this page isn't showing up in the toc on the dev site

[maximlt]

I've pushed a few changes, please let me know if you disagree with some of them, they can always be reverted.

My main comment, which applies to pretty much every user guide that attempts to demonstrate Datashader, is that the file size ends up being quite large, close to 30MB in this case. If you don't have a super fast connection, you can see the page being loaded slowly (this one takes so long for me https://holoviews.org/user_guide/Large_Data.html!), and there's also some time spent rendering the page and its plots that have many data points. At the same time, most of these plots would deserve to be inspected with a live Python kernel to see the full benefit of the applied approach (LTTB, datashader).
Instead of displaying the real Bokeh plots, wouldn't it be better if we displayed pretty images or GIFs? (I don't want to block this PR if we think that's what we should do, that could be done in a 2nd iteration).

[maximlt]

With the new set up the notebooks need to have a title.

[maximlt]

I'm not really for documenting unreleased features. But something we could do would be to document this as if HoloViews 1.19.0 was already released. In that case, we should also update the code so it's already able to accept and pass the new values. What do you think?

[droumis]

Ok, I included a note that these options are available starting in HoloViews 1.19.0. @hoxbro please be sure to include the tsdownsample PR in the next minor release. I don't think anything further needs to be updated in hvPlot code.

### Enhanced Downsampling Options

Starting in HoloViews version 1.19.0, integration with the [tsdownsample](https://github.com/predict-idlab/tsdownsample) library introduces enhanced downsampling functionality with the following methods, which will be accepted as inputs to `downsample` in hvPlot:

- **lttb**: Implements the Largest Triangle Three Buckets ([LTTB](https://github.com/predict-idlab/tsdownsample?tab=readme-ov-file#:~:text=performs%20the-,Largest%20Triangle%20Three%20Buckets,-algorithm)) algorithm, optimizing the selection of points to retain the visual shape of the data.
- **minmax**: For each segment of the data, this method retains the minimum and maximum values, ensuring that peaks and troughs are preserved.
- **minmax-lttb**: A hybrid approach that combines the minmax strategy with LTTB.
- **m4**: A [multi-step process](https://www.vldb.org/pvldb/vol7/p797-jugel.pdf) that leverages the min, max, first, and last values for each time segment.

[maximlt]

I don't think anything further needs to be updated in hvPlot code.

Not quite right I think, downsample can just be a boolean to use downsample1d, I don't think there is a way to customize how it's called from hvPlot at the moment (or if there is, it's not obvious and needs to be documented). I'll open an issue, but since multiple algorithms are documented here, it should be implemented before the release.

[maximlt]

Not sure why this page isn't showing up in the toc on the dev site

It's now showing up.

[droumis]

My main comment, which applies to pretty much every user guide that attempts to demonstrate Datashader, is that the file size ends up being quite large, close to 30MB in this case. If you don't have a super fast connection, you can see the page being loaded slowly (this one takes so long for me https://holoviews.org/user_guide/Large_Data.html!), and there's also some time spent rendering the page and its plots that have many data points. At the same time, most of these plots would deserve to be inspected with a live Python kernel to see the full benefit of the applied approach (LTTB, datashader).
Instead of displaying the real Bokeh plots, wouldn't it be better if we displayed pretty images or GIFs? (I don't want to block this PR if we think that's what we should do, that could be done in a 2nd iteration).

We decided in a meeting to not block this PR.. it would be nice in the future to have some mechanism to optionally facilitate images in docs rather than big plots

[droumis]

ready for merge?

[maximlt]

We decided in a meeting to not block this PR.. it would be nice in the future to have some mechanism to optionally facilitate images in docs rather than big plots

Yep, this sort of this is falling in our ever-growing bucket of nice-to-have-but-will-likely-never-happen! By the way, the Large Data guide in HoloViews does have a couple of GIFs, so adding GIFs to show what Datashader is capable of isn't unprecedented. I still think that would be the best thing to do to both make the page faster to load and better demonstrate hvPlot with Datashader, but as I don't really want to do it now, I'll just put that in the bucket too!

ready for merge?

Yes!