Golden files for every produced page

[iderex]

Depends on the plugin pages issue and on #60.

Once the pages are generated, the interesting question about a change is which
bytes it moved. A golden file per page turns that into a diff a reviewer reads
instead of a claim a reviewer trusts.

Each page is rendered from a fixture roster, not from the real one. A test that
judges against the live data proves the state of the data on the day it ran
rather than the behaviour of the generator, and it goes red for reasons that
have nothing to do with the change under review.

The fixture roster deliberately contains the awkward cases: a sentence with
markup characters in it, a plugin in each state, the longest and shortest names,
and a row with characters outside ASCII.

Updating a golden file is a deliberate act with the diff in the pull request
body, never a flag that regenerates them all quietly.

Done when

Every produced page has a golden file rendered from the fixture roster, a
one-word template change reds exactly the affected goldens, and the update path
requires an explicit command rather than being the default behaviour of the
suite.

[iderex]

The build writes three files and two of them are pages, so what a golden covers
has to be named rather than assumed:

go run . build
version 0.1.0
read templates/page.html.tmpl
read content/index.txt
read data/design-tokens.json (206 value(s))
wrote dist/index.html (1140 bytes)
wrote dist/privacy/index.html (3350 bytes, 2 checked, 3 promised, 2 residual)
wrote dist/.well-known/security.txt (379 bytes)
no assets/ in the tree, so nothing was copied verbatim
3 file(s) written into dist

Run 2026-08-12 at 39ab67a. Neither page is rendered from a roster. One takes
its words from content/index.txt, the other from content/privacy.txt, and
the third file is rendered from data/security-contact.json and is not a page
at all. So a golden set defined as every produced page has to say what each
page is rendered from: the fixture roster reaches the pages #28 produces and
reaches none of the three above.

The middle condition is the one that changes shape. Every page goes through one
template:

gh api repos/Flowfin/site/contents/templates --jq '[.[].name]'
["page.html.tmpl"]

Run 2026-08-12. A one-word change there moves every page at once, so it reds
every golden and the word exactly discriminates nothing. The change that does
discriminate is one to a single page's own input, reddening that page's golden
and no other, and that is the near miss the test is worth writing around. #73
is what makes the first case a property of one file rather than an accident, so
the two conditions are worth reading together.