fix(deploy-uni-hook): anchor price/balance/skew gates to the pool's own price

[aaronjmars]

What

Freeform balance / skew / "heavier-side" hooks generated by this skill compare the two virtual reserves. That comparison is a raw-price gate in disguise: StateLibrary gives amount0 = L*2^96/sqrtP and amount1 = L*sqrtP/2^96, so amount0/amount1 = 1/price and the liquidity L cancels exactly. The gate therefore tests the pool's raw price against an implicit 1.0, which only matches "balanced" for a same-decimals pair near parity.

Consequence: on any real pair the pool is permanently one-directional (one whole leg of trade reverts forever). A USDC(6d)/WETH(18d) pool sits ~8 orders of magnitude off; two 18-dec tokens at price 2.0 are already outside a 10% band. And "every trade rebalances toward 50/50" is false anyway (a full-range position is already 50/50 by value at any price).

The skill's own behavioral-test scaffold hides this: setUp() opens the pool at 1:1 (the one price where such a gate looks correct) and the generator may not edit setUp().

Changes

1. Fleet audit rules -> Gates rule 6 (SKILL.md): documents the L-cancels defect and the fix - snapshot the pool's own sqrtPriceX96 at afterInitialize and gate against that, never 1.0. Requires the afterInitialize callback (flag 0x1000) on any price/balance/skew hook.
2. Fleet audit rules -> Tests + step 4 build/test guidance: a price/balance/skew gate must be asserted off 1:1.
3. templates/Hook.t.sol: new _freshPoolAt(uint160 sqrtPriceX96) helper (fixed scaffolding) that repoints the test pool to a second pool on the same hook at an arbitrary price, so the generator can assert both legs away from parity without touching setUp().

No frontmatter or file-count change, so catalog/skills.json is unaffected (verified against the ci-skills-json normalized diff). forge fmt --check passes on the template.