[why]
When generating a monospaced (i.e. isFixedPitch) font we assume that all
glyphs have the same width, all glyphs expect the first three magic ones
(`.notdef` (ID0), `.null` (ID1), and `nonmarkingreturn` (ID2)).

Because `isFixedPitch` means that all glyphs have the same width it is
hardcoded that we just include the width of ID3 - all following glyphs
will get the same width. This is done to reduce the mtx table size.

There are two reaons why this is not correct.

a)

If the font comes with a glyph `NULL` at ID1 (which is allowed) we will
move the 'unexpected' `NULL` further on, and insert a new `.null`.

The font will end with this glyph order:
`.notdef` (ID0), `.null` (ID1), `nonmarkingreturn` (ID2), `NULL` (ID3), ...

But the width of `NULL` is zero, and so all width in the font will be
zero. Instead of hard-coding 'ID3' as first real width we would need to
check if it a glyph with a width.

Of course we could also correct out auto-inserted magic glyphs and just
keep the name `NULL` for ID1. Albeight that might also be desirable, it
is not a complete solution, because:

b)

With the 'new' (modern?) way to determine isFixedPitch (footnotes),
especially with allowing two width in a fixed pitch font (one "1 cell"
wide and one "2 cell" wide width), we need to check and possibly
encode all widths.

Example follows, InputMono-Regular.ttf is used with ttx.

Original situation before read in:

    <GlyphOrder>
      <GlyphID id="0" name=".notdef"/>
      <GlyphID id="1" name="NULL"/>
      <GlyphID id="2" name="nonmarkingreturn"/>
      <GlyphID id="3" name="space"/>
      <GlyphID id="4" name="A"/>

After we exported the font:

    <GlyphOrder>
      <GlyphID id="0" name=".notdef"/>
      <GlyphID id="1" name=".null"/>
      <GlyphID id="2" name="nonmarkingreturn"/>
      <GlyphID id="3" name="NULL"/>
      <GlyphID id="4" name="space"/>
      <GlyphID id="5" name="exclam"/>

Note that NULL is moved down to index 3 (and all following glyphs are
reordered).

The current code hard codes that ID3 shall be the last entry in mtx, and
`NULL` has a width of zero in the example.

Furthermore there are glyphs of zero width or the "2-width" width, that
also all will loose their values (originals following):

    <mtx name="caron" width="700" lsb="139"/>
    <mtx name="caroncmb" width="0" lsb="-211"/>
    <mtx name="caroncmb.salt" width="0" lsb="-114"/>
    <mtx name="ccaron" width="700" lsb="100"/>

Note the cmbs with a zero width and a negative lsb. Even if we correct
the `lasthwidth` to for example 4 we would loose all differentiation on
widths:

    <mtx name="caron" width="700" lsb="139"/>
    <mtx name="caroncmb" width="700" lsb="-211"/>
    <mtx name="caroncmb.salt" width="700" lsb="-114"/>
    <mtx name="ccaron" width="700" lsb="100"/>

[how]
We can not assume any width structure on a `isFixedPitch` font, and the
mtx table can not be shortened any more rigurous than usual, meaning we
walk backwards through the glyphs and find the last change in width,
dropping all numbers in the end that stayed the same.

`FigureFullMetricsEnd()` already find the optimally short end for the
mtx table while making sure to keep all widths.

[footnotes]
The 'modern' isFixedPitch is introduced with commit:

  c5eb85b `Modernize fixed pitch flag computation` (fontforge#5506)

The bug turned up in these issues:

ryanoasis/nerd-fonts#1948
https://gitlab.archlinux.org/archlinux/packaging/packages/ttf-input-nerd/-/issues/2

Signed-off-by: Fini Jastrow <ulf.fini.jastrow@desy.de>