ROHME-3: enable ROHM/NVDA and ROHM/GME LP tokens as bondable reserves

ROHME-3: enable ROHM/NVDA and ROHM/GME LP tokens as bondable reserves WHAT THIS ENABLES. Holders supply their own liquidity to a ROHM/<equity> Uniswap V2 pair and bond the resulting LP tokens to the Treasury for discounted ROHM on the standard 5-day vest. This is the classic Olympus LP-bond: it converts user-supplied liquidity into protocol-owned liquidity, which deepens the pool permanently. Deeper pools also directly raise how much fee revenue FeeCollector.harvest() can convert per call, since its ceiling is a fraction of pool depth. 1. setPriceFeed ROHM/NVDA LP 0xCB549c9C9E8cf3e52BF37B26Fc208AeD3b3f7D16 -> 0x749F443D3f1AEC830D47F767A4Dbb96912b2801a 2. enable RESERVETOKEN ROHM/NVDA LP 3. setPriceFeed ROHM/GME LP 0x744270585eDFfB51E9783574c0b95c4063BCF639 -> 0xBD746e0c56b8b7784c00552850778812A5565238 4. enable RESERVETOKEN ROHM/GME LP HOW THE LP IS VALUED. answer = min(vOther, sqrt(vRohm * vOther)) / lpTotalSupply, where the ROHM leg is valued at a flat $1.00 backing target and the paired leg at its Chainlink answer. The min() is the important part: the mark is capped at the value of the NON-ROHM leg, because that is the only side the Treasury actually recovers when it unwinds the LP. The protocol's own ROHM sitting in the pool is a liability, not backing. This is the same math UniV2PolLpFeed already uses for the genesis POL. It is up-manipulation-proof: k is swap-invariant, so no flash-loan swap can raise the mark. Donating to the pool does raise it, but a donation is a gift to every LP holder and so is not profitable to an attacker. A 20 pct haircut applies on top, and an F02 price band set at 4x fair value acts as a circuit breaker. AUDIT HISTORY - READ THIS. An earlier revision of these feeds was filed for registration and then withdrawn after an audit found two High-severity defects, both reproduced with proof-of-concept tests: - the pool was marked 2*sqrt(vRohm*vOther), i.e. as if BOTH legs were recoverable backing. The over-mark is 2*sqrt(backing/marketPrice) - 2x at par and worse below it. Measured end-to-end, a bond at the derived floor took $4.00 of real assets for 5.818 ROHM of liability: 31 pct instant dilution, while reported backing per ROHM went UP. - the ROHM leg read treasury.totalReserves()/baseSupply(). Because _auditReserves WRITES totalReserves from a loop that READS it back through the feed, the permissionless auditReserves() became a fixed-point ratchet on the mint budget - measured excessReserves 399.99e9 -> 420.99e9 with zero economic activity. Both are fixed in the feeds this proposal registers, and the fix is regression-tested against a re-implementation of the buggy version. WHY checkPaused IS FALSE ON THESE TWO REGISTRATIONS. A UniswapV2Pair does not implement oraclePaused(), and the Treasury treats a reverting staticcall on a flagged token as UNPRICEABLE - setting it true would brick the LP permanently. The ERC-8056 corporate-action gate is enforced inside the LP feed against the equity leg instead. RISKS. Both pools are EMPTY at filing, so the first liquidity provider sets the price ratio unilaterally; each pair should be seeded with a small honest amount before bonding opens. Suppliers take impermanent loss between ROHM and the equity. These LP feeds read the LastPriceFeed adapters from ROHME-2, so they inherit that proposal's weekend trade-off - if ROHME-2 does not pass, these LPs are unpriceable outside market hours.