fix: augment HL position reconstruction with clearinghouseState

[Flotapponnier]

Problem

HL userFills API is capped at 2000 fills per call. Active traders with more than 2000 fills have positions whose open fill is outside the cap window. These positions:

• Generate real HL funding (captured by userFunding which has no cap)
• Are invisible to reconstructHlPositions (no fill in the returned data)

This caused the Gains/dYdX/Paradex carry projection to be severely underestimated for heavy traders, while HL's actual funding ($834 in the example) looked disproportionately large vs the projected carry ($2.83).

Fix

Fetch clearinghouseState (current open positions) alongside fills and funding. After reconstructHlPositions, call augmentWithHlOpenPositions which adds any currently-open position not already captured by the fill reconstruction (keyed by coin:side).

These positions get a slice from cutoffMs → now, which is the same window used by userFunding. This makes the carry projection consistent with the actual HL funding data.

Applied to all 4 call sites of reconstructHlPositions (HL→Gains aToBSim, HL→Gains bToASim, generic aToBSim, generic bToASim).

Test plan

• HL wallet with many fills: Gains/dYdX funding projection should now be meaningfully higher (not near-$0) when trader has large open positions
• HL wallet with few fills (< 2000): no change (clearinghouseState positions were already in fill reconstruction)
• No regression on venues without HL