Wire verifier preview to emit hunt verdicts (pass/partial/fail → hunted/wounded/escaped)

[Hmbown]

Wire the verifier preview to emit hunt verdicts. Composes the v0.8.46 verifier-preview scope with the verdict vocabulary that ships in v0.8.45 (#2092).

Concept

The v0.8.46 milestone already scopes a verifier preview that auto-spawns a fresh-context, read-only, time-boxed sub-agent on claim-of-done and returns pass / partial / fail. After #2092, codewhale has a shared verdict vocabulary (hunting / hunted / wounded / escaped) and the data structure to hold it. This issue maps verifier verdicts onto hunt verdicts and routes them through the same code path.

Mapping

Verifier verdict	Hunt verdict	Action
pass	hunted	Trophy card written. Session may auto-close per config.
partial	wounded (with verifier reason text)	No trophy. Verdict logged. Hunt remains open; runtime suggests follow-ups.
fail	escaped (with verifier reason text)	No trophy. Verdict logged. Hunt closed.

/force-complete (already in verifier-preview scope) becomes /declare-hunted — overrides the verifier and writes the trophy. Audit-logged.

In scope (v0.8.46 increment)

• Verifier subagent returns a structured verdict (replaces ad-hoc string).
• Verdict flows through the HuntVerdict enum from Rename /goal → /hunt: quarry, verdict vocabulary, trophy card writer #2092.
• Trophy card writer (from Rename /goal → /hunt: quarry, verdict vocabulary, trophy card writer #2092) fires on pass/hunted only.
• /declare-hunted override audit-logged to ~/.deepseek/audit.log.
• /tasks "verified" badge replaced with the verdict glyph (one symbol per verdict variant).
• [verifier] enabled = true plus a new optional [verifier] verdict_policy = "hunt" (default; only mode shipped this cycle).

Out of scope (defer to v0.9.0 / #2094)

• Three-jurisdiction policy (strict / evidentiary / permissive).
• Judge as a separately-configurable LLM with its own model id and prompt.
• Trajectory-aware judging (judge sees full session, not just claim-of-done snapshot).

Acceptance

• A session with [verifier] enabled = true finishing a hunt produces a verdict; the verdict is the hunt verdict.
• pass → trophy card on disk. partial → wounded log. fail → escaped log. All three observed in tests.
• /declare-hunted overrides and audit-logs.
• Verdict glyph renders in /tasks.
• Existing verifier-preview tests pass unchanged or are updated alongside the verdict-mapping commit.

Closes / partially closes

• Closes Support Codex /goal long-running task mode #891 (codex /goal long-running task mode) — verdict gate completes the long-running loop.
• Closes Goal command should start actionable goals without a second follow-up turn #2050 (Goal command should start actionable goals without a second follow-up turn) — verdict gate eliminates the two-turn dance.
• Related: Rename /goal → /hunt: quarry, verdict vocabulary, trophy card writer #2092 (vocabulary), /hunt jurisdiction system: configurable LLM-as-judge with strict/evidentiary/permissive policies #2094 (full jurisdiction).

Composes with the verifier-preview surface already scoped in the v0.8.46 milestone description.