fix: HL funding projection accuracy

[Flotapponnier]

Bugs fixed

1. Math.max(0, cost) in computeHlFunding
The cross-sim was ignoring funding received (negative cost). If your Gains positions would have received funding on HL, the projection was showing $0 instead of a negative number. This made HL look artificially expensive. Now the projection is signed (positive = paid, negative = received).

2. PONS in coinsToFetch
When building the list of coins to fetch HL funding history for, we were including Gains-exclusive coins (e.g. PONS) which don't exist on HL. This wasted an API call and a slot in the 6-coin cap, potentially missing actual HL-comparable coins.

3. 6-coin cap too restrictive
Raised to 20. A wallet trading 7+ distinct coins would miss funding history for the extra ones.

4. Position reconstruction includes non-HL coins
reconstructGainsPositions now only receives positions for HL-available coins, matching the fee accounting logic.

5. fundingProjected flag not handling negative funding
hlFunding > 0.01 → Math.abs(hlFunding) > 0.01

Test result (before vs after)

	Before	After
HL funding projection	$189.29 (only paid)	$160.17 (signed net)
PONS in coin fetch	yes (wasted slot)	no
Coin limit	6	20