Cleanup of #3070 .

[jmcwilliams403]

The variable extraCurlyness has nothing to do with the width the character cell, which may cause problems in the e part in composite characters (ex: æ, œ) or in derived builds (superscript, boxed, etc.), so therefore the 'x'-coordinate:
(df.rightSB - OX - 0.5 * extraCurliness)
should instead be written as:
(df.rightSB - OX - 0.5 * (df.width / HalfUPM) * extraCurliness).
This also incidentally affects expanded width (and by extension Quasi-Proportional) but this should not really cause any problems (or it may actually be better this way) as it just subtracts a proportionally adjusted slice off the side relative to the change in width.

Also use old form when rounded e is used when the font is upright, since the upright form for the form added by #3070 appears to be untested. In the images below, both upright and italic use the rounded variant to demonstrate this, as the "straight" variant is unchanged.

Also, drop the code (specifically my code) for making reversed e (ɘ) resemble normal e in the rare "back-italics" (a trait of left­handed handwriting) since the text explanation in #3070 appears to make a good argument to have it be based on the actual writing direction of the Latin script (in cursive) when it's read back, which does not change just because a person writes it with their left hand (in other words, it does not magically become Arabic).

EeƎɘƎǝ ƏəӘәЭэ Ҽҽ øꬿœꭢᶕɚ œꭀᴔæꬱᴂ

Monospace:

Aile: