Fixture: JianYing 9.0.0 on Windows (encrypted mirrors beside a plaintext primary)

[Phelim2023315]

Adds a committed fixture, a regression test and a docs/version-support.md row
for a capture the docs currently describe from memory only.

The capture

A real 剪映专业版 (JianYing) 9.0.0 on Windows 10.0.19045 project, taken with
capcut fixture after the app had opened a draft this CLI wrote:

File	State
draft_info.json	plaintext JSON, 3485 bytes — the primary, still the tool-written bytes
draft_content.json	encrypted blob (app-rewritten)
template-2.tmp	encrypted blob (app-rewritten)
draft_meta_info.json	encrypted blob (app-rewritten)
Timelines/project.json, Timelines/469574b7-…/draft_content.json	layout + encrypted blob

Redacted with capcut fixture: no media, home paths / emails / device
identifiers stripped. Content is two text segments ("smoke-test",
"smoke-cta-default") from the tool's own smoke test.

Why it is worth committing

1. A plaintext tool-written primary is accepted on this build. The draft was
   written, closed, opened in the app, and the added text segment was present and
   editable. The app then encrypted the legacy mirrors and left the primary
   plaintext — so the CLI keeps reading the project through its canonical file and
   never writes into the encrypted mirrors.
2. The 6.0+ note about the primary file name does not hold here. On the same
   build an app-authored project is a plaintext draft_content.json with no
   draft_info.json or template-2.tmp at all. That folder is the owner's and is
   not committed, but the consequence is recorded in the docs row.
3. target_timerange.start is not guaranteed. The smoke-test segment
   carries {"duration":3000000} with no start and no speed; the app renders
   it normally and capcut segments -H printed NaN:00NaN / undefinedx for it.

What the test pins

• diagnose selects draft_info.json as canonical and reports 9.0.0.
• The three encrypted mirrors are reported as present-but-unparseable, and the
  readable primary still carries the timeline — a capture with encrypted siblings
  must not break canonical selection.
• version reads JianYing / lv / 9.0.0 off the primary.
• The current write-guard verdict (known-broken / refuse /
  beyond_known_range) is unchanged, so any future narrowing of the 6.0+ band is
  a deliberate edit with this test in front of it.

What this does NOT claim

The write guard is untouched on purpose. A committed fixture backs the storage
shape; it is not a round-trip claim for every JianYing 9.0.0 build, and this is
one machine. Narrowing the >= 6.0 refusal (or relaxing it for plaintext-primary
projects, the way the 11.4.0 macOS row is phrased) should be a maintainer
decision on top of this evidence — see #148.

Testing

npm test
npm run lint

test/jianying-9-0-0-windows.test.mjs: 4 tests, all passing.

[renezander030]

Thanks for contributing the capture and regression tests! This is a useful addition, and I agree with keeping the write guard unchanged. Once CI has run and passed, this looks ready to merge.