Correctly handle data values <= 0 on a log scale

[StevenCHowell]

This is a continuation from issue #5389, partially adressed by PR #5477. There persists an issue where negative data is not handled correctly. All data <= 0 should be discarded before generating the plot.

As is, if values = np.linspace(-0.1, 0.9), a JS error complains that it "could not set initial ranges", probably because log(n)forn<=0` is not defined.

[clairetang6]

Taking a quick look at this, the problem is definitely because the computed start for the range is NaN. I remember there was code (that I took out EDIT: this code is actually downstream in the log_ticker and was not taken out.) which was labeled hotfix, where the minimum data value was set to 1 whenever it was < 0. That fix also lead to weird behavior (flipped ranges) because the set minimum of 1 could be greater than all of the other data values. Especially in the log plotting context, it is impossible to guess at a minimum value.

The best thing to do would be to set the minimum data value to the the minimum value that is still > 0. But the min data value is saved into the bounds for each renderer and the ranges only use that and don't access all of the data values.

I want to try out a few things to see what would be the best solution. @bryevdv @StevenCHowell do you think it is worth it to first add back the hotfix? I could make that small PR tonight and then figure out a better way to fix it.

Also, the workaround now would be to supply a range start and end.

[StevenCHowell]

@clairetang6, for the hotfix, setting it to 1 will obviously create other problems (typically my data ranges from 0 to 0.3 ish). As you say, something more like the following would be more appropriate:

x_min = min(x[x > 0])

I don't think there is a rush to patch this, just to have to return to it later. Did you see issue #5550? Perhaps consider this as well while trying things as it is closely related.