fix: full signal renderer + schema alignment (entry_zone, invalidations, size_hint_pct, horizon)

[deluonchain]

what this fixes

SKILL.md — renderer was missing 4 of 7 blocks

the old SKILL.md was a pure API spec with no rendering instructions. the agent was only rendering verdict / score / confidence / summary / drivers / risks / mandate and skipping:

• signal stack (momentum / flow / structure / volatility / liquidity)
• regime & macro context (regime_label, regime_confidence, base_eco_pulse, macro_pulse)
• market snapshot (atr_pct_1h, pool_age_days, turnover_ratio_24h)
• comparables (pool_age_percentile, liquidity_tier)

new SKILL.md adds explicit rendering instructions with field-by-field output format, null-omission rules, and hard rules against skipping signal stack or context blocks.

references/mandate-fields.md — schema mismatches corrected

four bugs aligned to the live internal_oracle.ts output:

field	old spec	corrected spec
entry_zone	{ low, high } object	[low, high] array
invalidation	singular string	invalidations: string[] plural array
size_hint_pct	decimal e.g. 0.03	percentage e.g. 3.0, clamped [0.5, 10.0]
horizon keys	h4_to_d1, scalp, h1_to_h4	4h-24h, 1h-3h, 1h-4h etc.

horizon key table added for all regime labels.

backend patches still needed (bankr-fs internal_oracle.ts)

these are code changes required on the backend — not in this repo but documented here for tracking:

1. entry_zone: { low, high } → entry_zone: [low, high]
2. invalidation: string → invalidations: string[] (wrap in array, add liquidity condition)
3. size_hint_pct clamp [0.005, 0.1] → [0.5, 10.0], base sizes 0.05/0.03 → 5.0/3.0
4. horizonForRegime() key strings fixed to match mandate-fields.md table
5. sanitize() leak scrubber: out.mandate?.invalidation → ...(out.mandate?.invalidations ?? [])