Fix center of rotation with rotation_mode='anchor'

Remove ttconv backwards-compatibility code

Remove fallback code for glyph indices

This allows checking that there are no _new_ failures, without
committing the new figures to the repo until the branch is complete.

ci: Preload existing test images from text-overhaul-figures branch

Also, check some expected conditions at parse time instead of somewhere
during use of the data.

ci: Fix image preload with multiple conflicts

Add typing to AFM parser

As described in *TeX: the Program* by Don Knuth.

New font constants are set to the nearest integral multiples of 0.1 for
which numerators and denominators containing normal text do not have to
be shifted beyond their default shift amounts at font size 30 in display
and text styles. To better process superscripts and subscripts, the
x-height is now always calculated instead of being retrieved from the
font table (which was the case for Computer Modern); the affected font
constants have been changed.

A duplicate test was also fixed in the process.

At the call site
https://github.com/matplotlib/matplotlib/blob/51fbfc4eb0e882ef7e95ceab9777c7047f4db819/lib/matplotlib/_mathtext.py#L1706-L1709
the box should clearly go from `cur_v + off_v` to `cur_v + off_v -
rule_height` (this is why `cur_v` is shifted by `+ rule_height` just
before; also at that point some print debugging indicates that y's go
*downwards*), so `Rule.render` should indeed call `render_rect_filled`
from `y - h` to `y`.

Computer Modern values are taken from `cmsy10.tfm`, and divided by the
x-height in that output to match the scale used in Matplotlib.

DejaVu Sans/Serif and STIX constants are taken from the embedded TeX
table extracted with FontForge.

The text height includes both the ascender and the descender, but the
logic in _get_layout is that multiline texts should be treated as having
an ascent at least as large as "l" and a descent at least as large as
"p" (not a height at least as large as "lp" and a descent at least as
large as "p") to prevent lines from bumping into each other (see changes
to test_text/test_multiline, where the topmost superscript was close to
bumping into the "p" descender previously).

Fix confusion between text height and ascent in metrics calculations.

This mostly has a minimal change to results as the axis height is fairly
closely aligned with the minus sign as previously implemented.

These are mostly wider than the previous calculation of the 'm' width.
The values are invalid in the DejaVu fonts (-2048), so those still use
the old method.

…rics

mathtext: Fetch quad width & axis height from font metrics

Replace lookup table for font constants based on the family name
by methods in their respective classes.

Removes non-local call of private _get_font method in
_get_font_constant_set.

This simplifies implementing dynamically loaded font constants.

To replicate LaTeX behaviour, distinguish between "italic" and "normal"
math. In particular, digits should be set italic in the \mathit env.

For `cm`, use cmti font for "it"
For general UnicodeFont (stix, DejaVu, ...), maps digits to roman or
italic alphabet, depending on "normal" or "it" environment.

Ensure that \mathnormal is parsed and sets digits upright.

* Cache text rotation Affine2D

Fix tests

* Direct array creation

More robust BBox creation

* Speed up min/max calcs

* Fast path for unrotated text

* Faster FontProperties copy

* Faster array operations

* Faster shape check

* Code review updates

* More robust FontProperties hashing / copying

* Skip redundant wrapped text context manager

* Code review updates

* Restore stub __copy__

* Text consolidate rotation code path

* Prefer np.vstack().T to np.column_stack() for speed

Revert "Prefer np.vstack().T to np.column_stack() for speed"

This reverts commit 2e32436.

Simplify column stack

* Code review updates

* Cleanup

---------

Co-authored-by: Scott Shambaugh <scottshambaugh@users.noreply.github.com>

Apparently, when adding the additional fields in matplotlib#31050, I never noticed
that some of the required fields were not exposed.

Also, fix a few typos in the field names.

Consider a multiline Text object with nonzero rotation.
Text._get_layout performs the layouting of the individual lines.
Previously, it would return (among other things) the axes-aligned bbox
of the whole Text (which can be much bigger than the rotated bbox),
as well as the metrics of each individual line. mpl.text._get_layout
would then take these individual metrics and unrotate their positions
(to horizontal lines) to compute the size of the *rotated* bbox of the
text (and its position as well); this is used solely when drawing a
(text-aligned) bbox around the text (`plt.text(..., bbox=...)`).

The back-and-forth rotation seems a bit complicated; one can instead
simply return the relevant info in Text._get_layout as well: the rotated
bbox size is already available (in the variable previously named
`corners_horiz`), and getting its position is also simple.  Do that,
and get rid of mpl.text._get_textbox (and also adjust the callers of
Text._get_layout accordingly).  Also add some additional drive-by
reformatting of Text._get_layout.

Remove mpl.text._get_textbox.

TST: Switch mathtext tests to mpl20

ft2font: Read more entries from OS/2 font table

mathtext: add mathnormal and distinguish between normal and italic family