I believe this will fix MacOSX not displaying the icon. #30

The move from optparse to argparse caused a regression: positional
arguments could no longer intermix with optional arguments.  This fixes
that. (As of Python 3.7, parse_intermixed_args solves the problem,
but we want to support 2.6, maybe 2.5.)

Force the range to be -0.1 through 0.1. This is what matplotlib does
anyway, and this suppresses a warning.

Slightly simpler, smaller, and easier to style. As a bonus, you can
copy/paste them.

adding extra checks when variable data is loaded to prevent crashing and some cosmetic changes to html report output

added global and variable attributes to report pages; fixed bug in the io module that was causing global attributes to never be loaded; corrected the behavior for purple marking circles behind small numbers of mismatch points and added that behavior to some other kinds of mismatch plots

fixing variable sync in the gui and making a preliminary change to improve statistic display precision

If we had a file open in that "slot" (A or B) already, and the selected variable name is also present in the new file, pick that.
If that doesn't work, check to see if the selected variable name in the other file (B or A) is present, and if so pick that.
Otherwise stick with the existing default: the first item.

This should speed things up a little bit when bouncing between files, as the variable the user is most likely interested will already be selected.

In the GUI, if the user changes the A variable, and the A and B variables were the same, try to change the B variable to match.

Changed from catching activated to currentIndexChanged, as the former doesn't work when the selection is programmatically changed.

If the file is inaccessible, raise the same Exception the built-in global open() would.  This makes for more specific and helpful errors. This is done by doing an experimental open() on the file and not catching anything.

Old behavior was to raise an uncaught exception. Now it reports a warning and continues onward.

For variables who shape is "0" (that is, a simple scalar value), statistics calculations now work.  Although many of the statistics are silly, this is useful for report generation.

Implementation: DataObject has a new function "holding_array" which in most cases returns itself. But if self.data.shape==0, it returns a copy of itself with self.data's scalar value tossed into a single element array.  Now that the data is an array, as expected, the statistics code works fine.

"glance info --parsable" now outputs the results in tab deliminated format, repeating the filename on each line, making it easier to parse programmatically. Additional fields may be added later!

At the moment only supports "info".  May add support for other formats later.

In glance gui, if a variable has only a single value, it will be displayed directly in the "Data shape" field.

A 0 dimention numpy array is weird.  It's neither a scalar, nor a 1 dimensional array with a single element. "Display Data" silently refused to work for such values.

This change converts the 0d array into a 1d array with a single element immediately before trying to display it.  Result: Display Data now works on 0d arrays, showing the single value as expected.

Put sort directly into for loop; don't bother with giving it a variable.
Use named argument instead of relying on positional one.

Returns how many files could not be read, to a maximum of 255 (since the exit code is 8 bit).
This allows other processes calling "glance info" to detect errors.

Previously output was ordered in whatever order the set happened to offer them in. This was not stable between different Python versions.  Ordered alphabetically(ish) should be more human friendly, and will simplify the tests.