make trendlines more robust

[nicolaskruchten]

Closes #2549 (trendlines with zero slope) as well as a couple of other issues I found while testing:

• trendlines over timeseries data was just broken 🙈 https://stackoverflow.com/questions/61644142/plotly-how-to-show-trendline-for-time-series-data-using-plotly-express
• although we check for length > 1, we don't check for non-None/non-NaN length > 1
• we didn't have a lot of testing in general on trendlines/results, so I added a bunch

[emmanuelle]

Here if we have strings like the data column of px.data.stocks() this will raise an error message ValueError: could not convert string to float: '2018-01-01'. This is not a regression since the error message is better than the current one deep in statsmodels, but this line could be in a try except statement with an error message written by us.

[nicolaskruchten]

agreed, will fix