Translate the Japanese navigation command prefixes

[yasumorishima]

say-command still set the four prefixes to the English words, so a Japanese
synthesiser said them in English: "move 右", "zoom イン", "read 右",
"describe 右". They are now 移動 / ズーム / 読み上げ / 説明, the way ru, nb,
fr, hu and sv already translate theirs.

This is safe because of $CommandOffset (ec36e05): the offset is the length
of the English $NavCommand stem, not of the spoken prefix, so the suffix test
still matches. ja already carried the right offsets (5/5/5/9) — see #740 for
the two languages that do not.

Word order

Japanese puts the target before the verb, and the particle belongs to the
verb, so a $Particle is now set beside the prefix (に for move, を for read
and describe) and emitted inside each direction branch:

command	before	after
MoveNext	move 右	右 に 移動
ReadNext	read 右	右 を 読み上げ
ZoomIn	zoom イン	ズーム イン
ZoomInAll	zoomインを最大にしました	ズームインを最大にしました

Keeping the particle inside the branch matters: a command whose suffix matches
nothing (MoveCellUp, MoveStart, …) still falls back to the bare verb, the
way it does in en.

The Zoom commands keep the English order on purpose. ズーム + イン is the
ordinary loanword, and the concatenation that produces
ズームインを最大にしました needs the prefix first. The two suffix sets are
disjoint — Zoom only produces In/InAll/Out/OutAll and the others only
Next/Previous/Current/LineStart/LineEnd — so splitting the branch loses
nothing.

Two rules that would have contradicted it

Reviewing the change surfaced two other rules in the same file speaking the
same words in the other order, so they move too:

• current (ReadCurrent / DescribeCurrent) never goes through
  say-command; it said 読み上げ 現在 and now says 現在 を読み上げ.
• move-next-no-auto-zoom-at-edge-math used 右に for all three verbs, which
  gives read and describe the wrong particle. The direction and its particle
  are now part of each verb branch: 右に移動 / 右を読み上げ / 右を説明,
  followed by できません.

Checks

• audit-translations ja: untranslated text 3577 → 3574; missing rules 0,
  extra rules 0. Rule differences 28 → 30, both of them in navigate.yaml and
  both intended — diffing the audit output against ja shows exactly one
  "variable difference" (the added $Particle) and one "structure difference"
  (the reordered branch).
• Adds tests/Languages/ja/navigate.rs, the first navigation tests for ja.
  They assert the command prefix with starts_with, because what follows it
  comes from NavigationParts and is not what this touches.

Wiring the new test module in needs one line in tests/languages.rs, which is
the only change outside Rules/Languages/ja and tests/Languages/ja.

🤖 Generated with Claude Code

[yasumorishima]

The one red check here is not from this PR.

• Failing test: Languages::en::alphabets::cap_cyrillic, in Test (no-unsafe, Rules.zip) — job. The panic is 'RefCell already borrowed' at src/speech.rs:2799, i.e. re-entry while the full unicode table is being loaded, not a speech mismatch.
• The same tree passes that same job on my fork: success.
• This PR only touches Rules/Languages/ja/navigate.yaml, tests/Languages/ja/navigate.rs and one line of tests/languages.rs; it does not touch anything the Cyrillic alphabet test reads.
• I have seen the same job fail non-deterministically before, on 2026-08-29, where the two runs of one commit disagreed (one red, one green) with this identical panic.

I could not find an existing issue for it (searched RefCell already borrowed and cap_cyrillic). Happy to open one with these two runs as evidence if that would be useful — I did not want to file it uninvited.