Every screenshot is taken by a spec

[VeryComplexAndLongName]

First of the six, implemented. Based on proposals-simpler-for-the-user
(#417), not on main — the change it implements is proposed there.
Merge #417 first and this retargets itself.

openspec/specs/openspec-workbench/spec.md has required since it was
written that a documentation picture is captured from the running
product by a test. Twenty of twenty-six pictures predated the
requirement, most last touched on 22 August.

Seven pictures are now captured.
packages/server/e2e/documentation-screenshots.spec.ts drives one
fixture workspace and takes the command runner (with a real openspec show completed in it), the summary, the diff preview, the change
editor, the harness entry point, the template catalog, and the processes
view showing a journaled run. Every capture waits on the element that
makes the picture worth having, which is what makes a changed screen
fail here instead of producing a picture of the wrong thing — it did
exactly that twice while the spec was being written.

A check keeps it that way. scripts/check-screenshots.mjs fails on
any picture that neither a capture writes nor a dated entry accounts
for, and on an entry that outlives its picture. It reads the capture
sources rather than a list of pictures, so pipeline.spec.ts — added by
#414 while this was being written — is found without being named. Nine
tests, wired into lint and test.

Nine extension pictures are editor-native — tree views, context
menus and the editor itself — so they are listed with the date each was
actually taken (22 August), not with today's. A date claiming a retake
nobody performed would be the same defect this change removes. The
retake is a human-only item.

Three things worth a second opinion:

• The four 0*.png files are not captured. They are assets for a
  post published elsewhere and no document here references them.
  Capturing them would mean building fixtures for a permission request,
  an observation record and a failing mechanical check to keep four
  pictures fresh that nothing shows. They are listed under a third
  reason, published-asset, and tasks 2.8-2.11 are left open: the
  question is delete them or reference them, and it is not a lint's to
  settle.
• Three captures are not byte-stable between runs — they show the
  fixture's temporary path and a run timestamp. Measured, not assumed;
  recorded in the tasks rather than fixed by freezing fixtures every
  other spec shares.
• No capture publishes the machine it was taken on. The first run
  put C:\Users\<account>\AppData\Local\Temp\... into two pictures;
  those fields are now masked with a flat grey rather than cropped, so a
  reader sees that something was covered.

Browser suite: 17 tests, 17 passed, 3.2m. npm run verify: exit 0, 1998
tests across 157 files. openspec validate --strict --changes: 7
passed, 0 failed.

🤖 Generated with Claude Code

[VeryComplexAndLongName]

Reopening to trigger the workflow now the base is main.