Allow passing in a column to plot for error_bars

[weiji14]

The pygmt.fig.plot method has an error_bar (E) argument that isn't very usable right now. Ideally we would be able to pass in a list or numpy array of error values (e.g. standard deviation).

As agreed above, we'll figure out how to pass in an 'error' column with error_bars (E) and rename 'sizes' to 'size' in a separate PR next time.

Originally posted by @weiji14 in #666 (comment)

There's also a Python package called uncertainties at https://github.com/lebigot/uncertainties/ which wraps error values around numpy arrays, it might be cool if we can integrate that with pygmt somehow! I.e. pass in an uncertainty array to fig.plot() and have it plot the error bars directly. Also could tie this in with the pint package at https://github.com/hgrecco/pint which has a concept of "Measurements" that includes the error with numbers.

[weiji14]

Just as a follow because there's a related post on the forum (https://forum.generic-mapping-tools.org/t/pygmt-plot-errorbars), what we could do is to copy and modify these lines

pygmt/pygmt/src/plot.py

Lines 207 to 213 in 65b099a

	if "G" in kwargs and not isinstance(kwargs["G"], str):
	if kind != "vectors":
	raise GMTInvalidInput(
	"Can't use arrays for color if data is matrix or file."
	)
	extra_arrays.append(kwargs["G"])
	del kwargs["G"]

to use kwargs["E"] instead of kwargs["G"], or maybe turn it into a for-loop over ("G", "E"). That way, passing an array into error_bars (E) should work, but I haven't tested it.

[seisman]

Just a quick note. The order of the data columns is important. For error bars,

The x and/or y errors must be stored in the columns after the (x,y) pair [or (x,y,z) triplet]

Refer to https://docs.generic-mapping-tools.org/dev/plot#e for details.