[Implementer] Add hypothesis + kill-criterion regex gate to type:experiment clarifier and planner

Parent: #275 (audit item 2)

Background

Per the audit posted on #275 (item 2), the workflow ASKS for a falsifiable hypothesis + kill criterion in two places but does NOT statically enforce either:

1. .claude/skills/issue/clarifier.md:130-134 — clarifier asks the questions but advances status:proposed → status:planning based on subjective LLM judgement of "All clear" vs "Ambiguities remain". A type:experiment issue without a hypothesis can slip through if the clarifier-LLM deems it "minor".
2. .claude/skills/adversarial-planner/SKILL.md:43 — planner is instructed to include a Hypothesis section "if experiment". Same problem: instruction-not-gate. A planner that omits the section gets caught only if a Critic flags it, which is non-deterministic.

Proposed fix

A ~30 LOC static gate at each surface:

• Clarifier gate. Before posting <!-- epm:clarify v1 --> "All clear", regex the issue body for **Hypothesis** (or ### Hypothesis) AND **Kill criterion** (or ### Kill criterion) sections. If type:experiment and either is missing, post the questions unconditionally as ambiguities — do not advance.
• Planner gate. Before posting <!-- epm:plan v1 -->, regex the drafted plan body for the same headers. If type:experiment and either is missing, refuse to post; loop back to the planner with the missing-section feedback.

Acceptance criteria

• A type:experiment issue with no hypothesis section in the body cannot reach status:planning without explicit user override.
• A drafted plan with no Hypothesis section cannot reach status:plan-pending without explicit user override.
• Both gates can be bypassed with --force-hypothesis-skip for type:experiment issues that are pure ablations / re-runs (the clarifier already passes these without re-clarification).
• New tests in tests/test_skill_set_status_calls.py-shape: static fixture issue bodies hit the gate predictably.