fix(trust-pack): translate policy reasons, SIEM status, and OWASP note to ja

[killertcell428]

Why

ROADMAP.md names Japanese-language output as one of the two things that actually differentiate Aigis (the other is trust-pack itself — see #265). Reading the generated pack end to end found it wasn't fully true: three pieces stayed in English regardless of --lang ja.

What was broken

Section	Symptom	Cause
03 policy snapshot	All 16 default rules' "reason" column shown in English	aigis/policy.py's default_policy() hardcodes English reason strings; _policy_snapshot() never translated them
01 / 04 (x3 spots)	"SIEM forwarding: not detected (forwarders are configured in code...)" leaked verbatim into ja output	_detect_siem_forwarder() returned a pre-built English sentence, not a fact
02 control matrix	Weekly-report row's OWASP column read "— (cross-cutting; not an OWASP risk category)"	No ja variant existed for that field

Fix

• Added a rule-id → Japanese-reason dict in trust_pack.py, looked up by _policy_snapshot() for lang="ja". A rule id not in the dict (i.e. a user-authored rule) falls back to its own reason unchanged. The literal YAML further down the same section is deliberately left untranslated — it claims to be the actual file under version control, and translating it would make the doc lie about what's on disk.
• _detect_siem_forwarder() now returns bare facts (an env var value, a file path, or empty) instead of a finished sentence; new _siem_status_text(configured, detail, lang) renders the line per language. Applied at all 4 call sites (2 en, 2 ja — the en ones were already correct but now share the same function instead of a separate hardcoded string).
• Moved the OWASP explanation into what_it_does/what_it_does_ja (which already had a ja variant) and simplified owasp_llm to "—".

Also fixed (found while reading this code, unrelated but real)

• aigis.__version__ was "1.2.0", untouched since the v1.2.0 release (release: v1.2.0 — 4 new detectors + trust-pack/audit CLI rollup #228) — both v2.0.0 and v2.0.1 missed updating it while pyproject.toml moved on. This surfaced directly in aigis --version and in trust-pack's own "Aigis vX.Y.Z" line and the policy YAML's "Generated by" comment. Bumped to 2.0.1; added tests/test_version.py comparing it to pyproject.toml on every run so this can't silently drift again.
• save_policy() hardcoded "Generated by Aigis v0.3.0"; now reads __version__.
• Joined two lines in the audit-log-evidence template (en + ja) that were split mid-sentence in the source, producing an odd hard line break in the rendered Markdown.

docs/sample-trust-pack/

Regenerated end to end with the fixed code (aigis init --agent claude-code --signed-audit && aigis trust-pack --lang both --org "Example Corp（サンプル株式会社）" --contact security@example.com, plus the HTML build) so the committed sample matches what the tool produces today, not a snapshot from v1.2.0.

Test plan

• pytest tests/ — 1786 passed (was 1781; +5 in test_trust_pack.py, +1 test_version.py)
• ruff check / ruff format --check clean
• Regenerated pack manually scanned for the three known leak patterns — 0 hits in .ja.md files (still present, correctly, in the literal YAML block)