Optimize computeYT flow so that it's solvable on chain

[kinrezC]

Currently we're solving for the amount of SY to flash swap off chain and passing that as calldata to maximize the YT out in a swap. This is not operable because the block timestamp when the transaction will be processed in a block is unknown. We need to be able to submit an approximated amount of SY to flash swap, then if that approximated amount is within some epsilon_i wei of the root, we can go ahead and solve for the optimal SY up to some epsilon_j. If abs(epsilon_i) is too large the transaction should revert early because the root finding would be too expensive.