fix: fee-compare apples-to-apples accuracy (paginated funding, multi-slice, comparable notional)

[Flotapponnier]

Summary

• F1 — paginated HL funding history: fetchHlFundingHistory now fetches up to 5 pages × 500 entries, covering 180d+ windows (before it silently stopped at 500 entries / ~167 days)
• F4 — multi-slice position reconstruction: reconstructGainsPositions now emits one PositionSlice per size-change period per position instead of one slice using the final post-increase notional for the entire window; pre-increase periods now use the correct (smaller) notional
• F5 — comparable notional: tracks comparableNotionalUsdc (HL-listed coins only) and walletStats uses it for the Gains→HL sim, so the HL taker-equiv fee isn't inflated by PONS/Gains-exclusive trade volume; Gains-exclusive fees are also subtracted from the Gains actual total
• HL-comparable greying (from previous PRs feat: apples-to-apples fee comparison (HL-comparable coins only) #2215/fix: HL funding projection accuracy #2216): Gains-only pairs show "Gains only" badge and are greyed out in the trade table

What was wrong before

Issue	Old behaviour	After fix
180d HL funding	Collapsed to 0 (500-entry cap)	Paginated, up to ~833d coverage
Gains positions with increases	Used final (larger) notional for full window	Correct notional per size-change period
PONS/exclusive coins in HL equiv	Inflated HL notional by non-HL trades	Filters to HL-available coins only
Gains actual fees in comparison	Included fees on non-HL coins	Subtracts exclusive-coin fees

Test plan

• Gains vs HL with 180d window: HL funding projection should be non-zero
• Gains wallet with PONS trades: exclusive badge shown, excluded from comparison total
• Both comparison directions render correctly (A→B and B→A)
• TypeScript passes (npx tsc --noEmit)