Add withdrawFromStrategy and depositToStrategy methods to vault

[DanielVF]

Problem

Current the strategists primary tool is reallocate which can move funds from one strategy to another. (The strategist can also move all funds from a strategy, or have all non allocated funds allocated to their default strategies.) There are a few problems:

1. This is not always the most gas efficient or slippage way of moving funds. For example, if the strategist needs to move funds from Convex to both AAVE and Compound, this requires the gas expense of two withdraws from Convex, and possibly extra slippage.
2. When funds are directly held by the vault rather than by a strategy, these funds have to be allocated to their default strategy first, then reallocated, which results in the funds being deposited then withdrawn from the default strategy and finally deposits into the target strategy. It's 3x as much gas as the job actually needs.
3. The current reallocation setup is complicated to automate reallocation. If we can first withdraw from strategies/coin combos that have too much, and deposit to strategies/coin combos that need more, it gets much simpler.

Possible Solution

I'm thinking two new methods which would replace our vaultAdmin.rellocate():

withdrawFromStrategy(fromStrategy, [coins], [amounts]); - would move coins from one strategy to the vault.

depositToStrategy(toStrategy, [coins], [amounts]); - would move coins from the vault to one strategy.

[franckc]

WIP PR here: #1071

[sparrowDom]

Somewhat related to the discussion in this ticket, I was testing what would be the gains if we were to implement a feature for (meta) strategies to be able to reallocate funds between themselves without exchanging them back to stablecoin and rather leave them in 3crv and deploy that. Results are here:

To test the financial feasibility I ran 2 test cases on OUSD metapool:

• remove 10 LP from metapool, remove resulting 3pool LP into DAI, deploy back DAI to 3pool, deploy resulting LP into metapool.
• remove 10 LP from metapool, deploy resulting 3poolLP back to meteapool.

Loss in the first test case is 5.6 ‱ and in the second 4.4‱ with the difference of 1.25‱. Wanted to bring this data up in case anyone has strong opinions whether we should act on it. I think it is not a substantial enough financial gain to implement said functionality.