Fix: AMM fee rates must include the pool's LP fee - #7

The bug

An AMM swap pays the platform fee and the pool's LP fee. The pool reports them as separate lines, so the all-in rate is their sum — but venue_fees took platformBuy/platformSell alone, understating every AMM figure by the LP fee.

For our pools that's 3% instead of 5% on a buy. Consequences:

• The "acquired via AMM" break-even came out identical to the order-book one (both 3%), so the command's whole point — showing that where you bought changes your true cost — silently collapsed.
• A recovery ask priced off it sits ~2% too low, i.e. you'd take a "profit" that isn't one.

How it was found

Dogfooding. Running lofty account breakeven --margin 5 against a real account printed:

ACQUIRED VIA | ASK +5.0% | BREAK-EVEN ASK | BUY FEE % | TRUE COST
amm          | $71.37    | $67.97         | 3.00      | $65.59
book         | $71.37    | $67.97         | 3.00      | $65.59

Two identical rows with a 3.00% "AMM" fee is the tell — no unit test would have caught it, because the test fixture asserted whatever the code already did.

Verification — measured, not inferred

Live AMM quotes for the same pool:

This also matches the independent evidence already on record: a 1-token buy with totalSpent $11.90 debited $12.50 — exactly ×1.05.

After the fix, the same command reproduces the hand-derived numbers:

amm          | $57.12    | $54.40         | 5.00      | $52.50
book         | $71.37    | $67.97         | 3.00      | $65.59

Scope

• The order book keeps its own rates — the LP fee compensates AMM liquidity providers and has no order-book analogue, so mtBuyFeePct/mtSellFeePct stand alone.
• A pool with no lp line falls back to the platform rate (not a missing rate).
• A pool with no platform rate is still genuinely unknown and yields no scenario, rather than a wrong one.

Testing

New amm_rates_include_the_pools_lp_fee pins all three cases. The existing pool fixture was also corrected to the real observed shape ({lp: 2, platformBuy: 3, platformSell: 3.5}) instead of invented values — which flipped one assertion honestly: swapping out now costs more than resting an ask, so it needs a higher break-even, not a lower one.

All 60 tests, fmt, and clippy -D warnings clean; re-verified live after installing.

Figures in this description are illustrative, not account data.