Fixes #1095

previously in the matplotlib plots, we used the default behavior for text which is that matplotlib itself renders the text (the matplotlib output svg contains svg paths rather than text)

this is simple because then matplotlib knows the size taken by the text and can set the correct viewbox so we don't need any processing of the svg afterwards.

but the drawbacks are that:

• it makes the plots bigger (around 40% I would say)
• matplotlib is not as good as a browser at finding the right font for each glyph and rendering the text. at the moment we don't specify a font stack for matplotlib so some glyphs will be missing and replaced by empty boxes, and even if we do the result may be suboptimal. This is reported for emojis in Table Report unable to read emojis #1095 but it will also occur for many characters outside of the default font on the system where the report is generated -- eg in my machine the default matplotlib config uses dejavu which misses Chinese and Japanese characters.

the alternative, chosen in this PR, is to ask matplotlib to put the labels as text in the svg, in which case the application which displays it (in our case a web browser) is responsible for rendering it. the main drawback is that matplotlib does not know how much space the text will take, so we need to adjust the plot's viewbox and width and height with javascript after the page loads.

not directly related, but this pr also improves a bit how the text is truncated for right-to-left scripts, so that the ellipsis ("...") is displayed on the correct side of the text (in more cases). and it also normalizes whitespace so that we don't have line breaks in matplotlib labels

main

this PR