This can be accomplished quite neatly using the utility imagemagick under Ubuntu:
sudo apt install imagemagick
Using the screen captures you have extracted from Strava:
HR.png Graph showing heart rate during a training run.
splits.png Graph showing the kilometre splits for a training run
power.png Graph showing power estimates for a training run
The full command line to stitch the images together, resize the completed image and then export as a suitable sized, clear image is as follows:
convert HR.png splits.png -append intermediate.png && \
convert power.png intermediate.png +append composite.png && \
convert composite.png -resize 800x -quality 70 final.jpg && \
rm intermediate.png composite.png
Line 1: Create the file intermediate.png with HR.png and splits.png vertically stacked.
Line 2: Add the power.png and the intermediate.png to composite.png with horizontal stacking.
Line 3: Convert the file composite.png to a lower quality jpg file and resize to save file size while retaining clarity.
Line 4: Remove the intermediate files.
And the resulting file is perfect for AI analysis while remaining small enough to not suck up too much of the size allowance:

There may be an issue with longer runs. For Marathon efforts it will be necessary to alter some of the command line as follows, this demonstrates how to change the fields:
convert HR.png power.png -append intermediate.png && \
convert splits.png intermediate.png +append composite.png && \
convert composite.png -resize 800x -quality 70 final.jpg && \
rm intermediate.png composite.png
And this produces the following:

And you see now that the longer field now neatly balances out the 2 smaller fields...