Make System Events Private from the Runtime

[shawntabrizi]

We do not want to allow any Pallet to accidentally read from the events deposited into the runtime.

Doing so can have a huge impact on the PoV size of a block, and currently this is a very easy way to shoot yourself in the foot as a Parachain.

This PR removes the "easy" access to the storage item outside of tests and benchmarking.

polkadot companion: paritytech/polkadot#3712

[kianenigma]

writing a small note here to make it a new issue: This can actually be removed. While it is an offchain read with no PoV consequence or time limit, we can achieve the same by storing the previous phase in some offchain storage, and then checking the new phase here. Whenever previous_phase != Off && new_phase == Off we call unsigned::kill_ocw_solution::<T>();.

[gui1117]

As far as understand reading events is ok as long as it is done after the first removal in frame_system::initialize.
Because in this case it only reads the element stored by the runtime thus it shouldn't be needed in the PoV. But I may be wrong.

But this PR also solve this problem.

[gui1117]

code is good to me, but doc should be improved IMO, and maybe we can actually read events without leaking into PoV, as long as they were removed before, but I'm not sure and in general I would advise not to read them inside the consensus runtime code.

[shawntabrizi]

bot merge

[ghost]

Trying merge.

[jaysonmald35]

Pls help

[bkchr]

As far as understand reading events is ok as long as it is done after the first removal in frame_system::initialize.
Because in this case it only reads the element stored by the runtime thus it shouldn't be needed in the PoV. But I may be wrong.

But this PR also solve this problem.

This is right, reading from the overlay doesn't include anything into the PoV.

[gui1117]

Pls help

what is your question ?