Improve VolatilityOracle accuracy and manipulation resistance

[haydenshively]

Addresses

• Bandit - Uniswap Formula Drastically Underestimates Volatilty sherlock-audit/2023-10-aloe-judging#38
• Bandit - Uniswap Aggregated Fees Can be Increased at Close to Zero Cost sherlock-audit/2023-10-aloe-judging#45

Fix for 45

• Increase FEE_GROWTH_AVG_WINDOW from 6 hours to 72 hours, making it cost 12x more to manipulate. This measurement isn't precise enough to pick up intra-day variance in the first place (was just noise before), so the increase doesn't hurt anything.
• Reduce IV_CHANGE_PER_SECOND by a factor of 4
• Linearly interpolate between IV samples instead of discretely jumping, giving Borrowers a heads up as to where IV is headed. In case of manipulation, this forces the attacker to wait and hope that Borrowers don't respond. It also helps in normal operation, as Borrowers are less likely to become liquidatable all at once, reducing congestion and price impact.

Fix for 38
The issue is not as bad as the report would have one believe:

• Currently there's no way to short, but due to opportunity cost, users should still withdraw if they're not making money
• Soon there will be a way to short, using something like Panoptic
• Governance can adjust nSigma between 4 and 8 (from a default of 5), so it should be relatively easy to compensate for any systematic error.

That said, there's one simple improvement that I've implemented here -- an exponential moving average for IV, where the upwards gain is 5x larger than the downwards gain. This smooths out changes (similar to the linear interpolation described above) and empirically causes it to match the "true" IV more closely.

Note

One post-facto explanation for this is that when the price moves rapidly, sophisticated actors are the first to update their Uniswap positions, making our measurement more accurate for a brief period. Once the dumber capital catches up and creates passive positions in the new trading range, IV becomes artificially depressed again.