Take the reader that decodes a fax, which changes nothing here

[tannevaled]

reader v0.4.2 → v0.5.0. No code change.

This changes nothing this package produces, and the number is the point of
saying so.

Measured

Two binaries, each verified with go version -m:

old  github.com/go-pdfkit/reader v0.4.2   github.com/go-pdfkit/pdffont v0.3.0
new  github.com/go-pdfkit/reader v0.5.0   github.com/go-pdfkit/pdffont v0.3.0

For every one of the 2 268 forms in /Users/Shared/pdfforms the harness
dumped, and compared byte for byte:

• whether the document has a form at all, whether it carries XFA, whether it
  asks for its appearances to be redrawn, and its default appearance string;
• every field's whole name, kind, value, shown text, read-only / required flags,
  MaxLen, quadding, widget count, button states and option count;
• the appearance the file already carries, for every widget of every field;
• and a freshly drawn appearance for each fillable text field, filled with a
  probe string of the letters pdffont v0.3.0 taught the family to read
  (Příliš žluťoučký kůň Łódź Gdańsk Iğdır Șoseaua Ķīlis ąčęėįšųū Erdős),
  including the content stream byte for byte.

files compared	2 268
output differs	0
of the 1 633 real documents, files with a form	1 086
fields read	133 241 across all 2 268 (129 560 in the 1 633 real)
text fields filled by the probe (real only)	88 168
appearances drawn from those (real only)	88 159
appearances that differ	0

The 133 241 figure matches the corpus's own documented field count exactly, which
is the check that the harness was reading the whole thing.

Why it is taken anyway

A form does not carry its own scans, so there was nothing here for a fax decoder
to change. It is taken because this module resolves reader alongside four
siblings that do need v0.5.0 — and because a module of this family lagging its own
dependency is exactly how a defect gets shipped twice: pdffont shipped the
21-second reader.Open denial of service to its callers for one release for
precisely that reason.

Not measured

Wall clock and peak memory. Three other corpus jobs were running, so a timing
figure would be fiction, and there is no output change for one to explain.

Also here

A .gitignore. This was the only module of the family without one, which is how a
coverage profile nearly went in with the previous commit; its five siblings all
ignore exactly this file. (Landed in #1; noted here for completeness.)

Gates

go vet, gofmt, exact 100% statement coverage, nine cross-compile targets, all
pass locally with GOWORK=off CGO_ENABLED=0.