LaTeX with words in it

[tannevaled]

extract v0.2.0 fixes a defect that made the text this produces valid and
useless: a run's width, its size and the width of a space in it were worked out
without the text matrix's own scale, so a document that puts its scale there —
which TeX does in every PDF it has ever produced — had a word break inserted
between every pair of letters.

Measured over the first 400 real forms, converting each to LaTeX and counting
the words in the body that are one or two letters long:

before 58.15%
after 26.17%

The output is also 7.6% smaller for the same 400 documents, because a page's
worth of single letters takes more room than the words they were.

Figures are unchanged at 297 drawn and 829 given up on: those were settled by
taking the reader that decodes a fax, which turned 241 \framebox{unreadable
image} into \includegraphics one for one.

reader v0.6.0 comes with it, and with it the /Crypt filter that 209 files —
12.8% of the real forms — carry. This module's own test is what caught the
contract change underneath: a page whose content will not decode must still be
reported rather than turned into an empty document, which is now what
extract.Runs guarantees.