Convert axhline, axvline, and axline to layout shapes

[robertoffmoura]

mpl_to_plotly silently drops reference lines. axhline, axvline, and axline produce lines with blended or axes transforms, which the renderer skipped with "I found a line that didn't have 'data' coordinates! I will not draw it, but I did not crash.". The converted figure ends up with the reference lines missing completely.

Fix: reference lines are now drawn as layout line shapes:

• endpoints are mapped to data coordinates using ax.transData.inverted()
• added as go.layout.Shape(type="line") on the current subplot
• color, alpha, linewidth, and dash styles are preserved

Before: 0 shapes (lines dropped with a warning).
After: lines render as layout shapes spanning the axis limits.

Snippet to reproduce:

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import plotly.tools as tls

cases = {
    "axhline": lambda: plt.axhline(0.5, color="red", linestyle="--"),
    "axvline": lambda: plt.axvline(2.0, color="blue", linestyle="-"),
    "axline":  lambda: plt.axline((1, 0.5), slope=0.5, color="green", linestyle="-."),
}

for name, build in cases.items():
    fig, ax = plt.subplots()
    ax.set_xlim(0, 4)
    ax.set_ylim(0, 2)
    build()

    p = tls.mpl_to_plotly(fig)  # 0 shapes before this fix
    print(f"{name:10s} -> {len(p.layout.shapes)} shape(s)")
    p.write_image(f"{name}_plotly.png")

axhline	axline	axvline

[camdecoster]

Thanks for putting this together! There are a couple of issues to address, but it's an improvement. I left one suggestion about axis refs that isn't required, but it's worth looking at for this PR.

Could you please add a CHANGELOG entry?

[camdecoster]

This works as long as the user doesn't zoom/pan the chart. Once they do, the lines no longer span the chart. You could update these to use "x domain" and "y domain" for the proper lines.

[robertoffmoura]

I updated the horizontal and vertical lines to use domain, but the diagonal ones are trickier. If we use domain for the diagonal lines, we'd get incorrect behaviour when panning or zooming in only one dimension. A workaround for the diagonal lines is to scale them by some factor (e.g. by 100). Not ideal, but I don't think Plotly supports infinite diagonal lines. Do you have any suggestions?

[robertoffmoura]

Thanks! I added the changelog entry, addressed your suggestions, and left one comment regarding the zoom/pan issue for axline.